Surface movement detection and stability evaluation of a loose fine‐grained soil slope during reservoir operation: A case study in NE reservoir

Landslide in reservoirs imposes challenges to reservoir operation and dam safety management practices. The understanding of the landside mechanism during reservoir operation is crucial to landslide‐related risk migration. During the reservoir operation from 2018 to 2021, a massive landslide occurred with over 107 m3 in total volume on the bank of the NE reservoir. The surface movement characteristics before and after the occurrence of landslides in the NE reservoir in the region scale were detected and interpreted by Sentinel‐2 time series images. Experimental studies were conducted to investigate the geotechnical properties of the fine‐grained soil. The slope stability was evaluated for a typical slope profile considering the rising water level using the extended Bishop's simplified method, which was implemented in the code STAB‐UNSAT. It can be found that the landslide in the fine‐grained soil occurred simultaneously when the water level rose. The cumulative area of soil slope failure on the left bank of the NE reservoir increased continuously during the reservoir operation from 2018 to 2020, especially had a remarkable increment from August to October in 2019. The extended Bishop's simplified method provides a more rational method to evaluate the soil slope stability. The slope failure mechanism of the studied soil, that is, collapse–erosion–slide upon the rising reservoir water has been proposed.


| INTRODUCTION
During the reservoir operation from 2018 to 2021, a massive landslide occurred with over 10 7 m 3 in total volume on the bank of the NE reservoir.Despite the landslide in this reservoir did not cause loss of lives, the severe sediment siltation reduced the effective reservoir storage and resulted in a higher dam-break risk during operation.The reservoir water fluctuation is one of the trigger factors for landslides; however, the soil slope failure mechanism is complicated and needs to be investigated at the moment.Moreover, it is crucial to give insight into the soil slope failure mechanism for not only slope stability evaluation but also safe reservoir operation in the future.
Numerous researchers have investigated water-induced reservoir landslides and provided massive case studies and theoretical considerations on the triggering mechanism (e.g., Alonso & Pinyol, 2011;Chen et al., 2023;Jones et al., 1961;Schuster, 1979).Jones et al. (1961) concluded that approximately 50% of the landslides occurred during reservoir impoundment and around 30% presented during drawdown operation after investigating the reservoir landslides at the Grand Coulee dam from 1941 to 1953.Riemer (1992) found that 85% of the landslides occurred within 2 years of the impoundment period after the project completion in a survey of 60 reservoir landslides.By interpreting the monitored results of groundwater level and deformation before a catastrophic landslide, Paronuzzi et al. (2013) concluded that the landslides of rock with high permeability were attributed to the reduction in effective stress upon the rising water level.Song et al. (2015) conducted numerical modeling of landslides due to water level variation and found that the soil-water characteristic curve (SWCC), saturated coefficient of permeability, and the velocity of water level fluctuation are influencing factors for the soil slope failure in reservoirs.Chen et al. (2023) argued that the response mechanism of the landslide deposit in the reservoir during filling-drawdown cycles is dependent on the permeability of the deposit.On the basis of previous studies, the occurrence of landslide in reservoirs is influenced by both external factors (e.g., water level fluctuation, rainfall infiltration, and earthquake) and internal factors (e.g., coefficient of permeability and SWCC).
Field investigation and physical model tests are commonly used to investigate the soil slope failure mechanism.Compared with the physical model tests, the field investigation, for example, surface mapping, subsurface exploration, and field monitoring, involves the real boundary condition and provides valuable information related to slope failure (e.g., Li et al., 2020).Conventional field investigation approaches are time and costconsuming, remote sensing techniques have attracted more and more attention in recent studies of geohazards, for example, laser scanning (e.g., Francioni et al., 2014), unmanned aerial vehicle (e.g., Beregovoi et al., 2017), and satellite remote sensing (e.g., Yang et al., 2019).For example, Sentinel-2 provides high spatial and temporal resolution satellite data for disaster monitoring including surface movement of landslides (Phiri et al., 2020).The limit equilibrium method (LEM) and finite element method (FEM) are two widely accepted approaches for slope stability analysis.Based on the comparative studies on twodimensional slope stability analysis examples by Liu et al. (2015), the factor of safety (FOS) produced by LEM is in good agreement with and slightly lower than that by FEM.Thus, due to easy-to-operate and reliability, the LEM with an appropriate assumption of inter-slice forces is preferred especially in engineering practice.
In this study, the Sentinel-2 time series images were used to interpret the surface movement characteristics before and after the occurrence of landslides in the NE reservoir in the region scale first.Second, experimental studies were conducted to investigate the geotechnical properties of the fine-grained soil which constituted the landslide material in the NE reservoir.The grain size distribution (GSD), Atterberg's limits, SWCC, collapse potential, and effective shear strength parameters in saturated conditions were determined.Moreover, the slope stability was evaluated for a typical soil slope profile on the left bank of the reservoir considering the rising water level.The contribution of matric suction to the soil's shear strength in the unsaturated zone above the groundwater table (GWT) was considered.The extended Bishop's simplified method was implemented in the code STAB-UNSAT.Lastly, the failure mechanism of the fine-grained soil landslide in the NE reservoir was proposed.with a total storage of 64 million m 3 , which is mainly used for crop irrigation.The study area in the postlandslide Sentinel-2 image is shown in Figure 1a, which includes the landslide in the NE reservoir (e.g., the blue and red empty circles) and the reference point for data interpretation (e.g., the red filled triangle).On the basis of the field investigation (Figure 1b,c), it can be seen that the surficial soil with an extremely loose structure and fine particle size is susceptible to erosion by water and wind.The elevation of the study area ranges from 2400 to 2900 m a.s.l.This study area has a temperate zone desertification climate.The annual average precipitation and evaporation are 35 and 2752 mm, which indicates an extremely arid climate and a typically unsaturated state in the topsoil.Considering the precipitation characteristics, earthquake records and human activities in the study area, the variation in the water level during the reservoir operation is the possible triggering factor for the landslide.
The two Sentinel-2 satellites launched in 2015 and 2017 as part of the European Commission's Copernicus program, which are equipped with optoelectronic multispectral sensors for surveying and provide images with a high spatial resolution (e.g., 10 m multispectral).The two satellites allow repeated surveys every 5 days at the equator and every 2-3 days at middle latitudes.In this work, 34 Sentinel-2 images were downloaded.Of these images, there are four images in 2017, seven images in 2018, 10 images in 2019, seven images in 2020, and six images in 2021 (Table 1).These images cover the area of landslide in the NE reservoir and cloud free for the landslide surface.

| Surface movement detection using Sentinel-2 time series
The time series images of Sentinel-2 were used to detect surface movement of the soil slope before and after the presence of a landslide.Figure 2 shows three images in 2018, two images in 2019, and one image in 2020.It can be seen clearly that a landslide was initiated in 2018 and the area of slope failure extended in 2019 and 2020 (see the region in the red circle in Figure 2).By field observations on the landslide (e.g., Figure 1c), it can be found that the fresh slip surface had a significantly lighter color compared with the original slope surface.Thus, the change in the area with a lighter color than that in the surrounding area can be used to quantify the change in the slope failure area.Due to the presence of overlapped reservoir water body at the soil slope toe in the images, time series images are grouped into four periods starting from the moment with a low water level: (i) from February 16, 2017to August 5, 2018;(ii) from August 20, 2018to July 16, 2019;(iii) from August 15, 2019to July 25, 2020;and (iv) from August 24, 2020 to July 15, 2021.
To detect the surface movement, a false color composite of the Sentinel-2 image was used and interpreted.For example, the increase in landslide area can be manually identified in Figure 3, in which the area in white color and that in red color indicate the slip surface and the original exposed ground, respectively.The landslide area (marked by the red polygon in Figure 3) at the corresponding moment can be evaluated by the amount of slip surface pixel and the represented area at the ground of each pixel in the image using the software ENVI (e.g., Yang et al., 2019).
Table 2 summarizes part of the interpreted results of the landslide area on the left reservoir bank at the detected moment.It can be found that the area of soil slope failure increased from 2018 to 2020, especially had a remarkable increment from August to October in 2019.It should be noted that it is difficult to estimate slope surface displacement at the magnitude of a few centimeters by the Sentinel-2 images with the 10-m spatial resolution compared with other techniques, such as Synthetic Aperture Radar (SAR) technique.However, the interpretation method using Sentinel-2 images for surface movement detection can efficiently overcome the incoherence issue when using SAR images to detect the ground movement of a few centimeters per day (Stumpf et al., 2017;Wright, 2002).
Figure 4 depicts the increase in the cumulative area of slope failure on the left reservoir bank and the reservoir water level versus time.The moments marked from "A" to "F" in Figure 3 correspond to those detected by Sentinel-2 images in Figure 2. The three red arrows highlight the periods with the remarkable increase in the landslide area.It can be concluded that the landslide occurs simultaneously with the rising water level during the reservoir operation.Therefore, the following sections are devoted to investigating the failure mechanism of the soil slope upon the rising water level.

| GEOTECHNICAL PROPERTIES OF THE FINE-GRAINED SOIL
Figure 5 shows a typical soil strata profile based on the borehole log, which indicates the slip surface in the finegrained soil layer (i.e., the polyline in red) when the water level reaches the maximum value of 2488.04 m above sea level (a.s.l.) from 2018 to 2020.The maximum depth of the slip surface in the fine-grained soil is around 50 m.On the basis of the in situ permeability tests using the ring | 49 infiltrometer, the saturated coefficients of permeability of the fine-grained soil range from 3 × 10 −6 to 4 × 10 −6 m/s; the in situ permeability coefficient of the gravel ranges from 2 × 10 −4 to 6 × 10 −3 m/s.Thus, the gravel at the slope toe can be regarded as a relatively permeable layer.The rock underneath the fine-grained soil layer is assumed to be impervious in the current study.The geotechnical properties of the fine-grained soil need to be investigated to account for the effect of the hydromechanical properties of the soil on the slope stability upon the rising water level.

| GSD and Atterberg's limits
To determine the GSD curve of the fine-grained soil, two commonly used methods, namely, the mechanical sieving method (i.e., ASTM D6913/D6913M-17; ASTM International, 2017a) for particle size greater than 0.075 mm, and sedimentation technique (i.e., ASTM D7928-17; ASTM International, 2017b) for the fraction finer than 0.075 mm were used.Figure 6 and Table 3 show the complete GSD characteristics of the fine-grained soil.The specific gravity of the solid particles is determined by the water Pycnometer procedure (i.e., ASTM D854-14; ASTM International, 2014) as shown in Table 3 as well.Atterberg's limits (or consistency limits), namely, liquid limit and plastic limit were determined according to ASTM D4318-17e1 (ASTM International, 2017c).Consistency limits provide information on the soil states with respect to water content (Table 3); they can be regarded as the best empirical tool for a rapid assessment of soil behavior (e.g., plasticity index).On the basis of the plasticity chart of the Unified Soil Classification System (ASTM D2487-17; ASTM International, 2017d), the fine-grained soil can be classified as a lean clay (labeled as "CL").

| Soil-water characteristic curve
SWC-150 Fredlund SWCC device manufactured by GCTS Testing Systems was used to determine the SWCC of the studied soils.The pressure plate method uses overpressure to extract water out of the pore structure of the soil specimen until equilibrium is achieved between the moisture content in the specimen and the applied overpressure, which is called "axis translation technique."The setup of the SWCC device is shown in Figure 7.The maximum overpressure that can be applied to the soil specimen is limited by the air-entry value of the high-air-entry-value disk, for example, 1500 kPa in the current study.To simulate the effect of wetting during the rising reservoir water level on the soil hydraulic property, the SWCC in the wetting (or adsorption) path was measured.The test starts with the undisturbed specimen with natural water content (2.6%) and dry density (1.30 g/cm 3 ) prepared by the sampling ring (64.2 mm in inner diameter and 31.4 mm in height).The applied air pressure started at 350 kPa, and the air pressure of 300 kPa was applied until the equilibrium state was achieved.Subsequently, the air pressure of 250, 200, 150, 100, 50, and 0 kPa were applied.The adsorbed water volume at the applied air pressure can be calculated by the water level change in the volume tubes on both sides.When the change in water volume is less than 0.5 mL for a continuous 24-h, it is assumed that the equilibrium state is achieved.The water mass adsorbed by the specimen at each Variation of cumulative area of soil slope failure and reservoir water level versus time.
F I G U R E 5 A typical soil strata profile on the left reservoir bank with the observed slip surface of landslide.
F I G U R E 6 Grain size distribution curves of the fine-grained soil.
T A B L E 3 Grain size characteristics and Atterberg's limits of the studied soil.
equilibrium state was estimated by the water volume reduction in the volume tubes.The settlement of the specimen at each equilibrium state was obtained from the change in the dial gauge reading.After the test was completed, the specimen was removed from the cell, the volume, and the gravimetric water content of the specimen were determined.Thus, the gravimetric water content (w) and degree of saturation (S r ) at each equilibrium value of matric suction can be obtained from the volume-mass relationship in unsaturated soils, as shown in Table 4.The van Genuchten model (van Genuchten, 1980) was used to fit the SWCC measurements: where S r,res is the residual degree of saturation; θ s and θ r are saturated volumetric water content and residual volumetric water content, respectively; α vg , n vg , and m vg are the fitting parameters.For the fine-grained soil in this study, S r,res and θ r can be assumed to be zero.By the nonlinear regression, the fitting parameters for the van Genuchten model can be obtained, as shown in Figure 8.

| Collapse potential test
The characteristics of collapse potential are determined by the oedometer (i.e., ASTM D5333-03; ASTM International, 2003).The undisturbed soil specimens at their natural state were placed in an oedometer.The predetermined vertical stress was applied to the specimen and kept until the settlement became stable, for example, smaller than 0.01 mm per hour.The subsequent vertical stress levels were applied step by step, and inundation was applied to the specimen with the distilled-deionized water at the desired stress level, for example, 12.5, 25, 50, 100 kPa, and so forth.Table 5 shows the initial states of specimens and the collapse potential test results.Figure 9a depicts the variation of settlement with applied stress levels and inundation for all specimens.It can be seen that discrepancies exist in the settlement of the specimens before wetting.This may be attributed to the difference in the initial (or natural) state of the prepared specimens, including the initial dry density, initial water content and initial height (see Table 5).
According to ASTM D5333-03 (ASTM International, 2003), the collapse potential of the soil specimen where h f is the specimen height at a given stress level after wetting, mm; h i is the specimen height at a given stress level before wetting, mm; h 0 is the initial specimen height, mm.As shown in Figure 9b, the collapse potential shows a nonmonotonic variation with the stress level upon inundation and reaches the peak value when p s is 100 kPa.The collapse potential is remarkable (e.g., I c greater than 2.0) when the stress level (p s ) is less than 450 kPa.It is worth noting that the overburden pressure of the submerged fine-grained soil at the slope toe is around 500 kPa, when the reservoir water level reaches 2488.04 m a.s.l.(see Figure 5).Thus, the collapse-induced destruction in microstructure and the consequent deterioration in shear strength of the loose fine-grained soil is expected to be one of the triggering factors of soil slope failure.The collapse index, I e is 2.3%, which is referred to be the collapse potential at stress level of 200 kPa.Thus, the degree of soil collapse is moderate according to ASTM D5333-03 (ASTM International, 2003).F I G U R E 8 Soil-water characteristic curve (SWCC) of the studied soil.compression test The effective shear strength parameters in a saturated condition, that is, cohesion c′ and friction angle ϕ′, were determined by the CU triaxial compression tests.Four undisturbed specimens were prepared using the cutting ring, which had an inner diameter of 39.1 mm and a height of 80 mm.The dry density and water content of the specimens at natural state are listed in Table 6.Before CU triaxial compression test, each specimen was placed on a porous stone which was above and in contact with the surface of the distill-deionized water.The saturation of the specimen was confirmed when the mass of the specimen was kept constant.According to GB/T 50123-2019 (Ministry of Water Resources of the People's Republic of China, 2019), the specimen failure is characterized as the maximum deviator stress when the axial strain is smaller than 15%.
The CU triaxial compression test results for soil specimens at different confining pressures are shown in Table 6.On the basis of the linear regression on the plot of (σ 1 − σ 3 )/2 versus (σ 1 + σ 3 )/2, the effective shear strength parameters, that is, c′ and ϕ′ were determined as 19.5 kPa and 30.7°.

| STABILITY ANALYSIS OF SOIL SLOPE UPON RESERVOIR IMPOUNDMENT
4.1 | Evaluation of soil slope stability using extended Bishop's simplified method Bishop (1955) devised a slope stability analysis method (referred to as the "Bishop's simplified method"), which considered interslice normal forces with the interslice shear forces ignored (see Figure 10a).Compared with the Ordinary method (or Swedish method of slices) with all interslice forces ignored, Bishop's simplified method is  more realistic in practice, which requires to solve the nonlinear FOS equation: where c is the soil cohesion, ϕ is the soil friction angle, β is the slice base length, W is slice weight, and α is the slice base inclination.For Bishop's simplified method, the iterative procedure is used to solve the nonlinear Equation (3) starting with an initial guess of FOS.Chen (2003) developed the code STAB, which provided commonly used LEMs for soil slope stability analysis, including Bishop's simplified method.It remains one of the core techniques to determine the position of the critical slip surface with the minimum FOS for a slope stability analysis.In the code STAB, Chen (1992) proposed a random search method for determining the critical slip surface with the global minimum FOS.This is advantageous over commonly used search techniques, for example, grid and radius technique and entry and exit technique used in SLOPE/W (Seequent Limited, the Bentley Subsurface Company, 2023), in which the critical slip surface may not have the lowest FOS in the searched region.In the code STAB, the shear strength parameters (i.e., c and ϕ) in Equation (3) above and below the GWT can be experimentally determined in the conditions of natural state and saturated state, respectively.However, the spatial variation in shear strength parameters above the GWT cannot be accounted for.
To investigate the effect of the spatial variation in shear strength parameter on the soil slope stability, the contribution of negative pore water pressure above GWT to soil shear strength is considered in this study.Fredlund et al. (2012) summarized the distribution of pore water pressure in unsaturated zones for hydrostatic equilibrium and steady-state flow conditions (e.g., infiltration and evaporation).The factor f w is typically greater than unity for the steady-state flow upward boundary condition (e.g., evaporation) in the arid and semiarid regions, as shown in Figure 10b.The negative pore water pressure (or matric suction) can be written as where u w is the pore water pressure above the GWT, γ w is the unit weight of water, and D is the vertical distance from a given point to the GWT.For the preliminary investigation, the steady-state seepage condition and f w = 1.0 and 2.0 are assumed.Vanapalli et al. (1996) proposed a nonlinear model for predicting the shear strength of unsaturated soil (τ f ) with respect to matric suction using SWCC as a tool: ( ) where σ n is the normal stress on the slip surface, u a is the pore air pressure, and u w is the pore water pressure.If the F I G U R E 10 Schematic illustration for the extended Bishop's simplified method: (a) circular slip surface and soil slices and (b) distribution of pore water pressure above the groundwater table (GWT) for steady-state flow upward boundary condition ( f w greater than unity).
pore air pressure u a is assumed to be zero, submitting Equation (1) into Equation ( 5) results in Thus, for the steady-state condition, the extended Bishop's simplified method treats the shear resistance of slices in the saturated zone and those in the unsaturated zone differently.In Equation (3), Combining Equations ( 3), (4), and ( 7), the FOS can be computed.On the basis of the extended Bishop's simplified method, the code STAB has been extended as the code STAB-UNSAT, which can be used to evaluate soil slope stability in both saturated and unsaturated conditions.
Table 7 lists the model parameters used in slope stability analysis using the extended Bishop's simplified method.For the sake of safety, effective cohesion was ignored when calculating the FOS.According to NB/T 10512-2021 (National Energy Administration, 2021), the effective frictional angle for calculation is determined based on the CU test results.
Figure 11 shows the calculated results of FOS for the typical soil slope on the left reservoir bank in the two scenarios.It can be seen that the FOS decreases from 1.70 to 0.96 when the water level increases from 2461 to 2488.04 m a.s.l.when the parameter f w is 1.0.When the f w is 2.0, the corresponding FOS increases which is consistent with the previous studies.However, the predicted slip surfaces (dark blue lines) are significantly shallower than the observed slip surface (red line).It indicates that the assumption of a circular slip surface may be invalid in this scenario and a more complicated failure mechanism is expected.

| Failure mechanism of the loose finegrained soil slope upon the rising water level
The design normal storage water level is 2497 m a.s.l., which is around 9 m higher than the maximum value of 2488.04 m a.s.l. in 2019.Therefore, it is crucial to predict the soil slope stability upon further rising water level for safe reservoir operation.On the basis of the field investigation and laboratory test results, the slope failure mechanism of the fine-grained soil with extremely loose structure upon the rising water level can be characterized as the following three stages: (i) the soil at the slope toe collapses and loss its shear strength upon saturation due to the rising water level, (ii) the collapsed soil at the slope toe is eroded and even removed by the reservoir water, and (iii) the soil slope slides with a nearly planar slip surface.Following the collapse-erosion-slide failure mechanism, the landslide volume upon the rising water level can be predicted.Figure 12 shows the progressive failure of the postlandslide slope with planar slip surfaces following the progressive advance of the erosion at the slope toe.Due to the soil slope failure mechanism, densification and antierosion protection measures for the studied fine-grained soil are supposed to be conducted.

| CONCLUSIONS
The surface movement characteristics before and after the occurrence of landslides in the NE reservoir in the region scale were detected and interpreted by Sentinel-2 time series images.Experimental studies were conducted to investigate the geotechnical properties of the fine-grained soil.The slope stability was evaluated for a typical slope profile considering the rising water level using the extended Bishop's simplified method.The main conclusions on the landslide in the NE reservoir can be drawn as follows: (i) The Sentinel-2 time series images provide an efficient method to detect and analyze the surface movement of landslides or potentially unstable slopes.It can be found that the cumulative area of soil slope failure on the left bank of the NE reservoir increased continuously during the reservoir operation from 2018 to 2020, especially had a remarkable increment from August to October in 2019.The landslide in the fine-grained soil occurred simultaneously when the water level rose.| 57 drawn that a more complicated failure mechanism, that is, collapse-erosion-slide failure mechanism upon the rising reservoir water is expected.More numerical modeling work on the landslide in reservoirs considering the hydromechanical coupling effect is deserved in future work.

2
| SURFACE MOVEMENT DETECTION OF SOIL SLOPE DURING RESERVOIR OPERATION 2.1 | Study area and satellite image source NE reservoir is the pivotal engineering for NE River, which is located around 120 km southeast of Cele County, Xinjiang, China.The reservoir is an annual flow-regulating reservoir F I G U R E 1 The study area: (a) Sentinel-2 image; (b) and (c) photos showing the landslide in the thick fine-grained soil layer on the left reservoir bank.

F
I G U R E 9 The collapse characteristics of the studied soil: (a) collapse curves, that is, settlement versus stress level and (b) collapse potential at different stress levels upon wetting.
(ii) The collapse potential test results on the undisturbed specimens indicate that the fine-grained soil has an T A B L E 7 Model parameters of the fine-grained soil used in code STAB-UNSAT.Scenario Reservoir water level (m a.s.l.) Dry density, ρ d (g/cm 3 ) Effective cohesion for calculation, c′ c (kPa) Effective frictional angle for calculation, ϕ′ c (°) Parameter f w loose structure and is susceptible to collapse upon wetting.The extended Bishop's simplified method was developed, which considered the contribution of matric suction to the soil's shear strength in the unsaturated zone.The model was implemented and the code STAB-UNSAT was developed.The slope stability evaluation results using the extended Bishop's simplified method demonstrated that the rising reservoir water resulted in a reduction in the shear strength of the finegrained soil at the slope toe and triggered a landslide in the reservoir.(iii) The theories related to unsaturated soil mechanics, however, can partly explain the failure mechanism of the fine-grained soil slope.The tentative conclusion can be F I G U R E 11 The FOS calculated by extended Bishop's simplified method: (a) before reservoir impoundment and (b) water level at the maximum value in 2019, that is, 2488.04 m a.s.l.FOS, factor of safety.F I G U R E 12 Proposed failure mechanism for the landslide upon the rising water level.YIN ET AL.
List of Sentinel-2 images used in this study.
T A B L E 1 Collapse potential test results of the studied soil.
T A B L E 5Abbreviation: CP, collapse potential.
T A B L E 6 Consolidated-undrained triaxial compression test (TCT) results of the studied soil.