Dynamic deformation monitoring and scenario simulation of the Xiaomojiu landslide in the Jinsha River Basin, China

The Xiaomojiu landslide is a typical high-elevation, long-runout landslide located in the Jinsha River Corridor. In this study, sequential InSAR time-series estimation was used to calculate the line of sight (LOS) surface displacements with descending and ascending Sentinel-1 images, and it turned out that the cumulative LOS surface displacement of the landslide was up to −78.4 mm during the period from October 2017 to April 2021 with the maximum LOS surface displacement rate of −38.5 mm/year. The landslide body could be divided into five zones (A, B1, B2, B3, and C) according to its topographical characteristics together with the LOS surface displacement time series. Combining engineering geological characteristics, LOS cumulative surface displacements with site investigation suggest that the Xiaomojiu landslide is likely to be a precipitation-triggered ancient traction rock landslide at the accelerated deformation stage. A dynamic simulation of the Xiaomojiu landslide with the PFC3D software shows that it could take approximately 65 s for the Xiaomojiu landslide from start-up to acceleration to deceleration to build-up of a barrier lake, followed by a simulation from the barrier lake to outburst floods with the HEC-RAS software indicating that the maximum depth of the outburst floods could be 13.5 m (15%), 24.6 m (25%), 42.1 m (50%), and 50.3 m (75%) along Qinghai-Tibet Plateau Transportation Corridor (QTPTC). It is believed that the results of this study provide a reference for landslide prevention along the QTPTC and the Jinsha River.


Introduction
Landslide is the most widespread geohazard in the world (Froude and Petley 2018;Lin et al. 2019) and, due to global climate change and human activity, their frequency has greatly increased (Piciullo et al. 2018).This is especially the case in the Tibetan Plateau of China (Royden et al. 2008), where complex geological conditions exist such as active faults, frequent geohazards, and rapid geomorphological processes (Guo et al. 2018(Guo et al. , 2022)).The area is extremely prone to landslides, introducing great risks to construction projects, as was the case with the 2018 Baige landslides (Xu et al. 2018;Fan et al. 2019;Zhang et al. 2020a), the 2008 Tangjiashan landslide (Xu et al. 2013;Shi et al. 2015), the 2000 Yigong landslide (Lu et al. 2002), the Xiongba landslide (Yao et al. 2022), and the 1952 Jiaxi landslide (Zou et al. 2006).Since the Xiaomojiu landslide is a typical high-elevation, long-runout landslide located near the Baige landslide, its long-term dynamic monitoring, and risk assessment are vital (Jia et al. 2020;Xiong et al. 2020).
Interferometric Synthetic Aperture Radar (InSAR) has the advantage of being an all-weather, all-day system, unaffected by cloud or fog, in addition to offering a huge capacity for archived data, and an increasing number of scholars have applied it to investigate landslide motion.It has been demonstrated that InSAR can be used to map the spatio-temporal features of landslide motion (Tomás et al. 2014;Hu et al. 2020a;Xie et al. 2020) and then identify landslide bodies at high spatial and temporal resolutions (Hilley et al. 2004;Herrera et al. 2013;Bayer et al. 2017;Shi et al. 2019;Jin et al. 2022;Zhang et al. 2022).Based on Unmanned Aerial Vehicle SAR (UAVSAR) data, Hu et al. (2020b) captured the motion detail of the Slumgullion landslide.Chen et al. (2021) used InSAR time series and improved SAR pixel offset tracking methods to recover the surface processes and failure mode of the mininginduced landslide in Guizhou Province.Using the Cross Wavelet Transform (XWT) and Wavelet Coherence (WTC) wavelet tool package developed by Grinsted et al. (2004), Tomás et al. (2016) examined the relationship between landslide motion and seasonal precipitation.Liu et al. (2021) proposed a new method to process C-and L-band SAR images to acquire an almost 12-year multidimensional time-series, so as to analyze the kinematic evolution, displacement characteristics, driving factors, and creep behaviors of the Shadong landslide along the Jinsha River.It should be noted that InSAR can only provide landslide surface displacements and no information about landslide subsurface deformation is available.Several approaches have been proposed to address this.For instance, assumptions could be made such as a spatially uniform landslide rheology (Booth et al. 2013).Crosta et al. (2014) reported that boreholes were drilled, logged, and instrumented with a range of sensors (e.g., open-pipe piezometers, borehole wire extensometers, and in clinometric casings) to acquire water pressure and subsurface displacement rate.Song et al. (2021) combined InSAR and seismic noise techniques to determine the landslide mechanism from ground to subsurface as well as its volume.A recent study (Wang et al. 2023) presented a novel way to use fiber-optic strain sensing nerves to determine the shear displacements of critical interfaces in landslides.
The high-elevation, long-runout landslides located along great rivers often have enormous volumes and energy, which, if failed, could result in landslide dams, barrier lakes, and then landslidelake outburst floods (LLFs).When the landslide dam is formed, it intercepts water and sediments from the upstream, resulting in a rise of riverbed and water level in the upstream, and deficits of water and sediments in the downstream.When the water level increases, the landslide dam may break and the extreme discharge of the LLF often has devastating effects on the downstream Landslides 20 • (2023)   Original Paper population and infrastructure (e.g., Kondolf 1997;Fan et al. 2020).With the development of landslide numerical simulations, many scholars have used numerical simulation methods for landslide dynamic simulations (Yin et al. 2016) and flood evolution process simulations (Butt et al. 2013).For example, Liu et al. (2020) used numerical simulations to reproduce the geohazard chain process, Sarma et al. (2020) studied the dynamic process of the geohazard chain process with different triggering factors (e.g., earthquakes and rainfall), and Mao et al. (2020) conducted single landslide scenario analysis and analyzed its damaged areas.
The Xiaomojiu landslide is a typical high-elevation, long-runout landslide along the Jinsha River Corridor and the Qinghai-Tibet Plateau Transportation Project (QTPTP) is located approximately 50 km downstream of this landslide.If the landslide failed, it could be a significant risk to the QTPTP.Hence, it is vital to carry out dynamic monitoring and risk assessment for this specific landslide.
In this study, the surface displacements of the Xiaomojiu landslide were first calculated using sequential InSAR time-series estimation.The body of the Xiaomojiu landslide was zoned based on high-resolution optical remote sensing images, DSM, and InSAR results.Furthermore, engineering geological characteristics, LOS surface displacement, and site investigations were combined to reveal the surface evolution mechanism and failure mode of the landslide.Then, we examined the seasonal oscillations of the deformation using wavelet tools.Finally, based on Particle Flow Code Three-Dimensional (PFC3D) software (Itasca Consulting Group Inc. 2006), a 3D model was established for the Xiaomojiu landslide (Pastor et al. 2015).The landslide accumulation simulation was imported into the HEC-RAS software (Hydrologic Engineering Center 2012) as a barrier dam for flood evolution simulation (Rahman and Chaudhry 1998;Lauber and Hager 1998;Wang and Bowles 2006a, b;Awal et al. 2007;Aliparast 2009;Regmi et al. 2013;Akazawa et al. 2014), thus reproducing and predicting the whole process of the landslide to barrier lake to outburst flood disaster chain.

Study area
The Jinsha River lies on the border of Sichuan Province and Tibet Autonomous Region.Historically, the Jinsha River Basin and its surrounding areas have frequently been damaged by geohazards, including landslides, debris flows, and snow avalanches, but landslides in particular have caused serious harm to local residents.The Xiaomojiu landslide ( 31.12 • N 98.70 • E ) is located in Boluo Town, Jiangda County, Tibet, approximately 5 km upstream of the 2018 Baige Landslide in the Jinsha River (Fig. 1a).
The Xiaomojiu landslide is situated in eastern Tibetan Hengduan Mountain and valley zone of the Jinsha River Basin with a typical tectonic and erosion landform.The collision of Indian and Eurasian plates has resulted in a strong uplift of Tibetan Plateau, intensifying the erosional downcutting of rivers, while the simultaneous coupling of mountain uplift and canyon downcutting has increased topographic variations, leading to the formation of high and steep slopes.The Jinsha River, where this landslide is located, is a V-shaped valley with a deep cutting, high and steep banked slope, serious weathering, and denudation.The peak elevation at the main scarp of the landslide measures approximately 3655 m, the landslide toe (2895 m) is strongly affected by the lateral erosion of the Jinsha River, and the relative height difference between the toe and crown of the landslide is 760 m, and the overall slope of the landslide is about 30 ~ 35°. Figure 1b is taken from Google Earth while Fig. 1c shows a corresponding 1:50,000 geological map.The landslide area is located on the Jinsha tectonic mélange belt between the Changdu-Simao and Dege-Zhongdian landmasses, while the Jinsha River subduction-accretionary mélange zone was formed by the collision of Tethys system.This mélange zone forms the junction of different geological tectonic units, a typical tectonic suture zone, and a concentrated development area of special unfavorable geological bodies formed by orogeny.It is mainly composed of metamorphic ultramafic rock (chlorite schist, serpentine), metamorphic clastic rock (quartz-sericite schist, twomica quartz schist, phyllite, slate), and marble (Robertson 2004;Cao et al. 2015).To further understand the geological settings of the landslide, three boreholes were drilled into the landslide body (Fig. 1b, D01 (60 m), D02 (50 m), and D03 (40 m)) and the total core recovery exceeded 75%.
In Fig. 1d, the landslide body is roughly divided into three layers.The first layer, with an approximate depth of 3 m, comprises silty clay sandwiched with gravel.The silty clay is gray-brown, slightly wet and plastic, with medium strength and toughness.The gravel is subangular in shape, with a particle size of 1-3 cm and a content of about 30%.The parent rock composition is predominantly schist, which is severely affected by strong weathering, and the surface contains a small number of plant roots.The second layer, at a depth of approximately 47 m, is a loosely structured gravel with sizes ranging from 3 to 15 cm.The gravel content is approximately 65% with gravel ≤ 50 mm in size, and the gravel pieces are subangular in shape and the parent rock is mainly schist.The silty clay has the same characteristics as the first layer filled with gravel; the depth of the third layer is 10 m and it comprises the same properties as the upper two layers.The gravel size is generally 2-10 cm, the diameter of the block can exceed 30 cm, and the gravel content is around 55%.Its unique geological background makes the Xiaomojiu landslide a typical high-steep valley-shaped broken loose body structure, providing extremely favorable geomorphological and basic material conditions for the formation of landslides.
The study area is located in a highland climate region.Boluo Town has a cold climate with four distinct seasons, with an annual average temperature of around 7.5 • C ranging from −22 to 9 • C .The temporal distribution of precipitation is uneven, mainly during the period from June to September, and the distribution of precipitation in a valley is less than that over a mountain.The windward slope angle is less than that seen with a leeward slope, the sunny slope angle is less than that of a shady slope, and annual evaporation is three times greater than precipitation.Surface water in the study area is mainly supplied by precipitation, and groundwater is mainly tectonic fissure water and pore water.Tectonic fissure water is widely distributed in this area, and the main sources of replenishment are atmospheric rainfall and snowmelt.Pore water is mainly distributed on the surrounding mountain slopes, and the sources of replenishment include surface water, rainfall, and snowmelt.Landslides 20 • (2023)

Data
In this study, multi-source remote sensing images were used to investigate the Xiaomijiu landslide.SAR images were obtained from European Space Agency's Sentinel-1 satellite -Terrain Observation by Progressive Scans (Torres et al. 2012) in Interferometric Wide (IW) swath mode, with a spatial resolution of approximately 5 m in range and 20 m in azimuth.A total of 106 images were selected from the Sentinel-1 descending track, acquired between 8 October 2017 and 26 April 2021, and 103 images were selected from the ascending track, acquired between 15 October 2017 and 21 April 2021.The Shuttle Radar Topography Mission (SRTM) (Farr et al. 2007) Digital Elevation Model (DEM) with 30 m resolution (https:// earth explo rer.usgs.gov) was used to remove topographic contributions.The tropospheric delay correction data was carried out with the Generic Atmospheric Correction Online Service (GACOS, http:// www.gacos.net) (Yu et al. 2017(Yu et al. , 2018a, b), b).QuickBird optical images dated 24 February 2011, 18 July 2017, and 14 April 2019, along with Unmanned Aerial Vehicle (UAV) data acquired on 14 October 2020 and 12 July 2021, were applied to analyze geomorphic characteristics of the landslide.Monthly precipitation data during the period from October 2017 to April 2021 were downloaded from the Global Precipitation Measurement (GPM, https:// gpm.nasa.gov).

Methodology Sequential InSAR time-series estimation
The 26 January 2020 and 9 January 2020 images were selected as the common primary images for the ascending and descending tracks, respectively.GAMMA software (Wegmüller et al. 2016) was used to co-register from Single Look Complex (SLC) images to the common primary images.Co-registrated SLCs were used to generate interferograms by setting the maximum temporal (50 days) and spatial (200 m) baselines.The Minimum Cost Flow (MCF) (Pepe and Lanari 2006) method was used to unwrap the interferometric phase.
Previous studies (e.g., Lin and Wang 2018;Zhang et al. 2020b) often ignored tropospheric effects when studying individual landslides.However, in cases such as the Xiaomojiu landslide, complex topography, geological structure, climate, and stratified tropospheric delay can cause seasonal oscillations (Samsonov et al. 2014) which may be misinterpreted as rainfall-induced movements (Dong et al. 2019).In this study, Zenith Troposphere Delay (ZTD) maps corresponding to the SAR acquisitions were obtained via GACOS (http:// www.gacos.net) and then applied to correct tropospheric delays for each interferogram.By checking the interferograms after GACOS correction, it can be found that terrain-related and longwave tropospheric delays were obviously reduced in most cases.After obtaining the high-quality unwrapped interferograms, the LOS surface displacement time series can be estimated using the Small BAsline Subset (SBAS) InSAR method as follows (Eq.1).
where N and M are the number of Sentinel-1 SAR images and interferograms, respectively; i ( i = 1, 2, … N ) represents the range change on each SAR acquisition date ( 0 = 0 ) and unw i (i = 1, 2 … M) represents the unwrapped interferometric phase.Equation 2 can be used to estimate the deformation timeseries (Berardino et al. 2002): where V 1 is the measurement residual, L 1 denotes the unwrapped interferometric phase of the archived SAR data, and A 1 and P 1 are the design matrix and weight matrix, respectively.T represents the transposition of a matrix, X (1) is the estimation of the parameter X firstly, and Q X (1) is its cofactor matrix.
After the archived Sentinel-1 images are processed, new images can be updated for each track with the sequential estimation approach.L 2 is considered newly added unwrapped interferograms, A 2 and B represent design matrixes, P 2 represents a weight matrix, and X (2) and γ are an estimation of the parameter (Eq.3).Based on the principle of Least Squares (LS) Bayesian estimation (Eq.4) (Yang 1991): (1) (2) According to Eqs. 2, 3, and 4, Eq. 5 can be deduced (Huang 1992): where stands for the updated deformation time-series, γ represents the cumulative deformation of updated Sentinel-1 SAR by date, Q [X ( 2) ; ] denotes their cofactor matrixes, and J x and Q J represent the gain matrix and updated cofactor matrix, respectively (Wang et al. 2019a, b).It should be noted that the maximum temporal and spatial baselines were set to 50 days and 200 m, respectively, in this study.In Fig. 2, the blue points denote the archived SAR images while the red points indicate the new SAR images used for updating the time-series.When a new Sentinel-1 image is acquired, the deformation parameters shall be updated.Finally, the updated LOS surface displacement rate and surface displacement time-series can be obtained.

Wavelet tools
Previous studies reported that precipitation and residual stratified tropospheric delays could lead to seasonal oscillations in InSAR surface displacement time series (Song et al. 2021).In this study, we used wavelet tools including Continuous Wavelet Transform (CWT), Cross Wavelet Transform (XWT), and Wavelet Coherence (WTC) to analyze InSAR time-series data of the Xiaomojiu landslide.The CWT is often used for analyzing individual time-series, and the result of the CWT is usually represented as a two-dimensional (2-D) image where two axes are defined by the frequency or period of the time instant and the time pattern.Thus, the locations of high values in the 2-D CWT representation indicate the presence of periodicities at particular times.If the relationship between two different phenomena needs to be analyzed, two separate CWTs can be combined by using the XWT or the WTC tools.The XWT is a 2-D representation of a complex number, obtained by multiplying the CWT of one time-series by the complex conjugate of the CWT of the second time-series; compared with XWT, WTC represents coherence between two results of CWT, which is calculated by normalized cross-correlation in the 2-D time-frequency domain.These wavelet tools have proven to be powerful (Grinsted et al. 2004;Tomás et al. 2016) and MATLAB codes have been made freely available (NOC 2014).Firstly, the non-linear LOS surface displacement component of P1 was acquired by subtracting its linear fitting values, computed using a linear least-squared fitting from the InSAR timeseries.The differential Zenith Tropospheric Delays (dZTD) were (4) taken from GACOS and the precipitation from GPM.We interpolated the time interval of the time-series of precipitation and dZTD to a time interval equally spaced to that of the LOS surface displacement (12 days).CWT can identify localized intermittent periodicities by analyzing time-series records (LOS surface displacements, dZTD, and precipitation) in time-frequency space.Then, the results of CWT for two groups (LOS surface displacements and dZTD, and LOS surface displacements and precipitation) were combined using the WTC and XWT tools.Finally, the seasonal variations of the Xiaomojiu landslide were examined in detail based on the results of wavelet tools (Tomás et al. 2014(Tomás et al. , 2016)).

Simulation of the landslide-barrier lake-outburst flood disaster chain
According to data taken from the three boreholes, it is speculated that the deformed body has a thickness of 90-100 m.Based on the PFC3D software, the discrete element method was used to build a landslide simulation model utilizing topographical data (Wang et al. 2003;Wu et al. 2009).It should be noted that four assumptions are made in the PFC3D software: (1) the sliding surface and the particle unit of the landslide are rigid, and the adjacent particles can be point contacts, (2) the contact model is a flexible contact with some overlaps in the contact points, (3) all the overlaps are smaller than the particle diameter, and (4) the bonding characteristics can be established by the contact between particles (Itasca Consulting Group Inc 2006).Note that high precision 3D topographical data with a spatial resolution of 0.17 m was acquired with UAV and employed in the landslide simulation in this study.Newton's second law of motion and law of force-displacement were used to update the position of particles, described as the motion of a particle (please refer to Itasca Consulting Group Inc (2006) for greater detail).Note that the Xiaomojiu landslide is an ancient landslide with part of its structure damaged during the initial sliding process, but this simulation did not consider this fact for the sake of simplicity.

Original Paper
According to the results of the landslide simulation, the barrier dam formed by the landslide would have an elevation of 2940 ~ 3000 m.In this paper, the hydrological analysis module of ArcGIS was used to calculate the inundation extent in the upstream of the barrier dam.When the final inundation elevation of the dam was set to 2940 m, the final reservoir capacity would be 4.13 × 10 9 m 3 .The hydrometric data from Batang station suggests that the average annual flow rate in the upper basin of the Jinsha River is 957.3 m 3 /s and it would take around 50 days for the barrier lake to spread.
The simulated landslide accumulation was thereafter imported into the HEC-RAS software as a barrier dam for flood evolution simulation.The Navier-Stokes equation was used for the calculation of two-dimensional non-constant flows in the HEC-RAS software; assuming that fluid is incompressible, the equation for the conservation of mass (continuity) is: The momentum equation is: where H represents water surface elevation, t is time, h denotes water depth, V is water velocity, and q and g are sink flow and gravitational acceleration, respectively.C f denotes the roughness rate of the river bed bottom, f is the Coriolis factor, k represents the unit vector in the vertical direction, and n is roughness (Zhou et al. 2016;Liu et al. 2020;Jia et al. 2021;Zhuang et al. 2022).

LOS average annual rates and cumulative surface displacements
Based on the sequential InSAR time-series estimation, LOS surface displacement rate maps were obtained using Sentinel-1 ascending and descending data during the period from October 2017 to April 2021 (Fig. 3).It is clear in Fig. 3 that the main scarp, main body, and toe of the landslide are deforming, with several obvious deforming areas are mostly concentrated in the main body and toe.The maximum LOS surface displacement rate in Fig. 3a is −38.5 mm/year while the LOS surface displacement rates in Fig. 3b range from −15.2 to 23.8 mm/year.Different patterns can be clearly observed in Fig. 3a and b and the ascending track is more sensitive to deformation than the descending track.This is because the SAR geometries for ascending and descending are different and the main body of the landslide is moving downslope and to the NE direction.
To further understand the evolution of this landslide during the study period, LOS surface displacement time series was examined in detail in Fig. 4. On 11 July 2018, the upper (south-west) and lower (north-east) edges of the landslide started to deform and, over time, the deformation magnitude and spatial extent of the landslide continued to expand.The minimum surface displacement of the upper edge decreased from −8 to −59 mm, the minimum surface displacement of the lower edge decreased from −5 to −72 mm, and the deforming area extended by 1.2 km 2 .Finally, the LOS surface displacement reached at −78.4 mm in the main body on 26 April 2021.In Fig. 4b, four deforming areas can be observed on the lower edge, main body, and upper edge of the landslide.In the northwest of the landslide, there is a strip-like deforming area, with the LOS surface displacement reaching −38.4 mm on 21 April 2021.Meanwhile, the deformation magnitude in the lower part of the landslide arose, with the LOS surface displacement reaching at 46.7 mm on 21 April 2021.

Interpretation of optical remote sensing images
Figure 5 shows five optical remote sensing images acquired at different times, among which Fig. 5a, b, and c were taken from the QuickBird satellite and Fig. 5d and e from UAV images.By combining the LOS surface displacements of the landslide with the optical remote sensing images and DSMs, it can be derived that the Xiaomojiu landslide has an armchair shape, indicating that one or more large gradient sliding events had occurred during its geological history.Firstly, we divided the Xiaomojiu landslide into three sections: the first zone (A) contains the upper part and sidewalls of the landslide; the second zone (B) is composed of the dislocation deforming area (B1), compressional deforming area (B2), and stress concentration area (B3); the third zone is at the lower edge of the landslide (C) next to the Jinsha River (Fig. 5f).Zone B is the main accumulation body of this ancient landslide.Comparing the optical images from 2011 to 2021, it can be seen that different zones of the landslide exhibited different deforming patterns.In 2020, a small dirt road (Fig. 5d) was constructed across the middle part of the landslide, cracks on the back wall of the landslide were enlarged, and the vegetation coverage decreased.Moreover, additional obvious cracks appeared in areas D1 and D3, the boundary of D2 showed signs of expansion, and the toe of the landslide (C) significantly widened.

Spatiotemporal evolution of the landslide
In Fig. 6, LOS surface displacements taken from ascending and descending images were superimposed onto the DSM of the landslide.Eight representative surface displacements were selected.The position of profile OI is shown in Fig. 6a and LOS surface displacement time-series along profile OI were extracted to analyze the spatiotemporal evolution of the Xiaomojiu landslide.Figure 6c and  d show the surface displacements along profile OI of the ascending Finally, based on DSM, high precision optical images, and InSAR derived surface displacements, the boundary can be determined and the landslide can be divided into five zones (Fig. 5f).In Fig. 6c  and d, zone A is on the back wall of the landslide and the minimum cumulative surface displacements are −15.0mm (Fig. 6c) and −12.2 mm (Fig. 6d).The head of zone B1, particularly the upper part next to the contact between the displaced material and the main scarp, has developed a platform, and the minimum surface displacements are −47.0mm (Fig. 6c) and −20.5 mm (Fig. 6d).In zones B2 and B3 (Fig. 6c), profile OI crosses cracks D1 and D3, and the slope becomes steeper, with a minimum surface displacement of −70.1 mm and −61.2 mm, respectively.When profile OI crosses zone C, the minimum surface displacement reaches −44.6 mm.In Fig. 6d, the minimum surface displacement of zone B2 is −20.2 mm while the maximum surface displacements of zones B3 and C are 34.7 mm and 44.8 mm, respectively.

Discussion
Engineering geological characteristics of the landslide Three boreholes were deployed on the slope to explore the engineering geological characteristics of the Xiaomojiu landslide.The positions of the boreholes are shown in Fig. 1b and the total core recovery and stratigraphy distribution of the three boreholes are shown in Fig. 1d.Based on the borehole data, an engineering geological profile was formulated (Fig. 7a).It should be noted that the bedrock of this landslide was not found during this field survey and thus the bedrock orientation is not included in the profile.From the engineering geological profile, it is evident that the five important deformation zones correspond to the locations set out in Fig. 7b.The cracks and scarps are mainly distributed in zones A and B1.By analyzing the stratigraphy distribution and InSAR surface displacements, the main potential sliding surface of the landslide (red line) can be estimated, the location of which lies between the position of scarp 1 and the toe of the landslide (Fig. 7a).According to the water content of the stratigraphy and static groundwater level, the flow direction of groundwater and water rich area is mainly concentrated in borehole 1 (38 m).The UAV survey and field investigation of the Xiaomijiu landslide were carried out on 14 October 2020 and 12 July 2021 (Fig. 7c).There is a lot of silty clay sandwiched with gravel on the surface of the landslide, and the gravel is subangular in shape, with a particle size of about 3 cm.Zone A is the source area of the Xiaomojiu landslide, where the ancient landslide slid down from the original ridge forming a chair-shaped landform.The total length of the back wall of the landslide is approximately 2500 m, and the average height difference is about 100 m with a maximum of 190 m.The upper area of the landslide source area exhibits limited vegetation coverage, a clear boundary, and strong rock fissures.The source area of the landslide formed gullies under surface diffuse scouring, causing the localized collapse of ridges in the northern area.The back wall of the landslide has undergone severe weathering, the bedrock in the back wall is exposed with limited vegetation cover, and there is a lot of metamorphic clastic rock and marble with fragmented rock.It can be inferred from this that the back wall of the landslide (and even the entire landslide source area) will continue to loosen and become more fragmented as a result of prolonged weathering alterations (Fig. 7c: A).
In zone B, a lot of chlorites schist and serpentine are scattered here, and lower density weeds are distributed.The main elevation of zone B1 measures from 3320 to 3460 m, with a slope of 22 • , a longitudinal length of 320 m, a lateral width of 280 ~ 680 m, and a circumference of approximately 1900 m.It is an arc-shaped stepped ridge and several gullies have developed along the slope direction,

Original Paper
of which the maximum width reaches 40 m and the deepest reaches 14 m (Fig. 7c: B1).The main elevation of zone B2 ranges from 3120 to 3320 m, with a vertical length of 380 m and a horizontal width of 1000 m.The rock deformation characteristics of this area are obvious due to the strong extrusion of the backslide downward, the rock structure being extremely fragmented and strongly weathered, and the extrusion and significant wrinkling phenomenon being quite significant.Localized collapse, rockfall, and surface weathering debris from the upper and middle parts of the landslide have collected in the wash, further destroying the integrity of the slope in this area.There is a secondary landslide in this area, with a length of 240 m, a width of approximately 310 m, a height difference of 130 m, and a steep wall at the lower edge of 12-22 m.Due to road excavation, a slippery surface has appeared in this area and there are tension cracks on the south-western edge of zone D1; on the one hand, this provides a route by which rainfalls can infiltrate the slope and, on the other hand, it destroys the structure of the slope (Fig. 7c: D1).
Zones B3 and C are mainly distributed at an elevation of 2900 ~ 3120 m, with a slope of 33 • (although locally greater than 40 • ), a longitudinal length of about 450 m, and a lateral width of around 1150 m.During the deformation process of the landslide, its lower edge locally slid down and squeezed the Jinsha River channel.The lower edge of the slope body is scouring and eroding at the Jinsha River, causing a local slump at the toe of the slope, the maximum height of which is approximately 90 m (Fig. 7c: C).The abovementioned engineering geological characteristics suggest that the Xiaomijiu landslide is an ancient rock landslide, i.e., at least one large sliding event occurred in history forming the Xiaomijiu landslide.
Fig. 8 LOS cumulative surface displacements of three selected points (P1, P2, and P3 in Fig. 6a) from ascending (a) and descending (b) tracks, overlaid with monthly precipitation obtained from GPM.Note that grey areas denote rainy seasons (June-September)

Seasonal oscillations in InSAR surface displacement time-series
The LOS surface displacements of the three points (P1, P2, and P3) marked in Fig. 6a were extracted.Figure 8a and b show the LOS surface displacements of ascending and descending data, respectively.The time-series of points were compared with monthly precipitation data downloaded from GPM in order to examine their relationships.As the rainy season in this region is mainly concentrated between June and September (shadows in Fig. 8a and b), surface displacements tended to accelerate during the latter part of the rainy seasons in 2018, 2019, and 2020.In particular, there was a significant acceleration in the three-point time-series following the 2019 rainy season (Fig. 8a), with maximum surface placements of −47.6 mm, −69.4 mm, and −54.8 mm for P1, P2, and P3, respectively.In Fig. 8b, the maximum LOS cumulative surface displacements of P1, P2, and P3 were −17.9 mm, −23.9 mm, and 40.1 mm, respectively.Deformation trends for the descending and ascending data were similar, with active responses to precipitation.
In Fig. 9a, the residual LOS displacement time series (indicated by the red line) was acquired by subtracting its linear component from InSAR derived LOS surface displacement time series, the differential Zenith Total Delays (dZTD, indicated by the black line) were taken from the GACOS, and the precipitation time series (indicated by the indigo line) was taken from the GPM.It is evident that the residual LOS displacement time series has cyclical fluctuations.To verify whether the deformation had seasonal oscillations, CWT, XWT, and WTC were used to analyze the LOS surface displacement time series for point P1. Figure 9b, c, and d show CWT of LOS surface displacements, precipitation, and dZTD.Strong power signals are evident with an annual (365 days) cycle since 2018 in Fig. 9b, and a clear annual cycle can be observed during the whole observation period in Fig. 9c and d.In addition, obvious power signals with a period of 1 to 2 months can be observed in Fig. 9b, c, and d.
Figure 9e and g show WTC and XWT relationships outside of the 5% significance level between LOS surface displacement and precipitation.In the shadow of Fig. 9f and h, the relationships of WTC and XWT between LOS surface displacement and dZTD also lie outside of the 5% significance level.In Fig. 9e and g, the LOS surface displacements and precipitation show a high wavelet coherence (Fig. 9e) and a high common power with a phase shift of about 20° in an annual cycle during the whole observation period (Fig. 9g), because the seasonal landslide oscillations often start shortly after the commencement of the rainy season.Therefore, the surface displacement of the landslide usually begins more than half a month earlier than the arrival of the precipitation peak (Song et al. 2021).

Original Paper
The WTC and XWT show that the LOS surface displacement and precipitation were in phase.Additionally, a common power with a period varying from 256 to 60 days can be observed in Fig. 9g.In Fig. 9f and h, the LOS surface displacements and dZTD show lower wavelet coherence in an annual cycle during the whole observation period than the WTC between the LOS surface displacements and precipitation (Fig. 9f).The WTC and XWT show that the LOS surface displacements and dZTD were in phase, but the relationship between the annual period of the LOS surface displacements and dZTD had an offset of about 40°.The LOS surface displacements appeared to have a stronger correlation with precipitation than dZTD.It can thus be concluded that seasonal precipitation played a more vital role than tropospheric delays for the seasonal oscillations of the LOS cumulative surface displacements.

Type of the Xiaomojiu landslide
Our site investigation revealed a large amount of gravel distributed on the slope surface.Numerous cracks were evident in the landslide's main body, especially in zones D1 (Fig. 10c and d) and D3 (Fig. 10b).Meanwhile, spring water was observed in zone D1 (Fig. 10a).These phenomena inevitably aggravate the deformation process of the landslide.With the LOS surface displacement timeseries from October 2017 to April 2021 being updated dynamically, the Xiaomojiu landslide was found to have been in an accelerated deformation state (Figs. 4 and 8), with no sign of ceasing, since June 2018.During the rainy season, the existence of numerous cracks and much gravel and clay on the slope itself intensifies the process of rainwater penetrating the landslide's main body, leading to acceleration of the deformation of zone B. This zone is dominated by weak, medium-thick layered limestone and dolomites with a high degree of weathering.Zone C is not only under pressure from zone B but is also washed by the Jinsha River water.In the toe of the landslide, many schist rocks and silty clay sandwiched with gravel can be observed, containing a few plant roots.Zone A is being pulled by zones B and C, and thus is also undergoing deformation.In summary, the Xiaomojiu landslide is now a traction ancient rock landslide, with seasonal precipitation playing a very important role in the landslide motion.

Landslide dynamic simulation
Based on the results of InSAR technology, visual interpretation of high-resolution optical remote sensing images, and site investigation, the landslide deformation characteristics and topographic and geomorphological features were obtained.The landslide was then modeled using high precision 3D topographical data acquired with UAV; a dynamic simulation of the Xiaomojiu landslide was obtained with the PFC3D software.Figure 11 shows the whole process of landslide movement.The Xiaomojiu landslide took approximately 65 s from start-up, acceleration, deceleration, and build-up.At the beginning of the movement, the slide produced significant displacement at around 4 s, resulting in instability in the landslide.During the 12 ~ 26 s period, landslide motion started to accelerate and the gravitational potential energy was converted into kinetic energy, at which point the slide started to block the Jinsha River channel.Compression of the lower edge by the upper edge of the slide resulted in the middle and lower edges having slightly greater velocity than the upper edge, with the lower edge of the slide entering the river first and reaching peak velocity.The subsequent 26 ~ 64 s period was the deceleration stage, when the slide passed through the Jinsha River and the lower edge was blocked by the opposite bank of the mountain.It then began to decelerate and piled up on the opposite bank due to energy consumption by tumbling and collision.The upper edge of the landslide underwent acceleration and deceleration from 26 to 64 s, after which velocity reduced sharply due to the blockage of the lower edge.At 64 s, most of the particles ceased moving.It can be seen that after the landslide failures, there is a great possibility of blocking the river and forming a landslide dam and then a barrier lake.

Simulation of the barrier lake to outburst flood disaster chain
To further analyze the risk posed by the Xiaomojiu landslide to the surrounding environment and the QTPTC, a simulation of the barrier lake to outburst flood disaster chain was conducted, based on the HEC-RAS software.Figure 12 illustrates the impact of the four simulated scenarios ( 15%, 25%, 50%, and75% ) of barrier lake failure on the QTPTC and the Jinsha River channel, and different percentages represent different degrees of barrier lake failure.In Fig. 12, blue denotes slow water flow rates while red means higher water flow rates.The flow velocity of the 15% breached flood is minimal, ranging from 2.0 to 4.0 m/s in the flat channel, with a maximum flow velocity of approximately 15.9 m/s.The flow velocity of the 25% breached flood is slow, with the maximum flow velocity reaching around 16.7 m/s near the villages of Guomai and Rangwong, and 7.2 m/s at the river bend.The distribution of flow velocities for the 50% breach flood is generally consistent with that for the 75% breach, being greatest near the villages of Guomai and Rangwong (17.3 m/s and 19.3 m/s, respectively).Figure 13 shows the distribution map of flood inundation areas along the QTPTC under the four simulated scenarios.The average depth of flood in the inundation area is 6.1 m (15%), 15.4 m (25%), 28.8 m (50%), and 35.7 m (75%), and the maximum depth of flood is 13.5 m (15%), 24.6 m (25%), 42.1 m (50%), and 50.3 m (75%) in the vicinity of Guomai-Lieba (Fig. 13b).Although the degrees of flood breakouts are different, their flow velocity distribution characteristics are similar.Flow velocities are higher at the bends of the river course and in the narrow area of the river valley, and lower near the two banks.In short, when floods occur, the water velocities and levels in the areas near the QTPTC are relatively high, which will have a significant impact on the construction and maintenance of the QTPTC.

Conclusions
In this paper, LOS surface displacement rates and time series were acquired using sequential InSAR time-series estimation.Then, the wavelet tools (CWT, XWT, and WTC) were employed to analyze the relationships between the LOS surface displacement and precipitation/dZTD.Finally, we performed numerical simulations of the landslide to barrier lake to outburst flood disaster chain.Several conclusions can be acquired as follows: 1. Combining high-resolution optical and DSM with the spatiotemporal evolution of LOS surface displacements, the landslide can be divided into five zones (A, B1, B2, B3, and C) Original Paper (Fig. 5f).The engineering geological characteristics of the landslide suggested that it is likely to be a traction ancient rock landslide with a 2500 m long back wall as well as a secondary landslide in zone B2. 2. The landslide is still in a state of accelerated deformation, and precipitation is an important driving factor.The minimum LOS surface displacement reached at −78.4 mm in the main body of the landslide during the period from 8 October 2017 to 26 April 2021.3. When the landslide fails, the Xiaomojiu landslide will take approximately 65 s from start-up to acceleration to deceleration to build-up of a barrier lake.The length of the landslide accumulation area is approximately 2023 m with the width of 900 m and the maximum height of 149 m.After the landslide blocks the Jinsha River, a barrier lake with an elevation of 2940 m and a storage capacity of 4.13 × 10 9 m 3 can be formed.
If the barrier lake fails, it will pose a huge risk to the QTPTC.
As such, it is urgently critical to monitor the landslide in realtime, and our team is planning to deploy a landslide early warning system (EWS) as demonstrated in Dai et al. (2020) in place to improve our capability of landslide disaster prevention and mitigation, which is believed to directly benefit to the construction and operation of the QTPTC.

Fig. 1 a
Fig. 1 a Location of the Xiaomijiu landslide (indicated by the green pentagon) with the coverage of Sentinel-1 images.Purple and blue rectangles represent the spatial coverage of Sentinel-1 ascending and descending images, respectively; black lines are provincial boundaries, pale blue lines denote rivers, and the red line indicates the QTPTC, b Google Earth optical image.The yellow line outlines the boundary of the Xiaomijiu landslide, and black dotted lines indicate the main scarp, main body, and toe of the landslide; D01, D02, and D03 (yellow dots) are borehole locations.c Geological map of the landslide.d Total core recovery and stratigraphy of three boreholes

Fig. 2
Fig. 2 Temporal and spatial baselines of Sentinel-1 images from ascending (a) and descending (b) tracks.Note that blue points denote the archived SAR images involved in the first round of time series analysis, and red points denote newly acquired SAR images involved in the following rounds

Fig. 3
Fig. 3 Sentinel-1 LOS surface displacement rate maps from ascending (a) and descending (b) tracks.Red color represents deformation moving away from satellite and blue color represents deformation moving towards the satellite

Fig. 6
Fig. 6 LOS surface displacement maps of the Xiaomojiu landslide from ascending (a) and descending (b) tracks.LOS surface displacement time-series are shown ascending (c) and descending (d) along with profile OI (indicated by black lines in a and b).Note that (1) P1, P2, and P3 in a and b show the locations of three selected points for further analysis in Fig. 8; (2) in a and b, red lines indicate zone A, pale blue lines denote zone B, and yellow lines imply zone C; and (3) triangles in c and d denote the dividing lines of the five deforming zones (A, B1, B2, B3, and C)

Fig. 7 a
Fig. 7 a Engineering geological profile of the Xiaomojiu landslide.The upper panel of a shows the distribution of five deformation zones, pale blue area indicates water rich area, pale blue line represents groundwater flow direction, red line denotes landslide sliding surface, and vertical black arrow represents precipitation.b denotes five deformation zones in hill shade.c shows field pictures of the deforming areas (A, B1, D1, D3, D2, and C)

Fig. 9 a
Fig. 9 a Residual LOS surface displacement time series (indicated by the red line) which subtracted the linear component from InSAR derived LOS surface displacements, dZTD (indicated by the black line) and monthly precipitation (indicated by the indigo line).b, c, and d represent CWT of residual LOS surface displacements, precipitation, and dZTD, respectively.Shadow designates a 5% significance level against noise.The lighter shadow denotes cone of influence by potential edge effects.e and g denote WTC and XWT between residual LOS surface displacements and precipitation, respectively.f and h represent WTC and XWT between residual LOS surface displacements and dZTD, respectively.Black arrows indicate relative phase shift, with the same phase pointing to right (0°).In e and f, yellow and blue denote the high and low wavelet coherences of two time series, respectively.In b, c, d, g, and h, yellow and blue represent the long and short time patterns common in the time series, respectively

Fig. 10
Fig. 10 Field pictures.a Spring in zone D1; b crack in zone D3; c and d cracks in zone D1

Fig. 12
Fig. 12 Distribution of flood velocities at different levels of barrier lake failure: a 15% , b 25% , c 50% , and d 75% .The red line is the QTPTC, blue color denotes slow water flow rates, and red color higher water flow rates

Fig. 13 a
Fig. 13 a The distribution map of flood inundation areas around the QTPTC under the four simulated scenarios (15%, 25%, 50%, and 75%) and b the water levels of the four simulated scenarios in the Guomai-Lieba