Chasing a Complete Understanding of the Yanshangou Landslide in the Baihetan Reservoir Area

Abstract

The Yanshangou landslide, located in the Baihetan Reservoir area, poses severe potential threats to the normal operation of the reservoir due to its distinct deformation characteristics and high sensitivity to reservoir water level fluctuations. This study systematically investigates the geological background, deformation characteristics, stability evolution, and landslide-induced surge hazards of the Yanshangou landslide in the Baihetan Reservoir area. This work only considers the influence of reservoir water level fluctuations, which is the dominant factor controlling the current progressive deformation of the landslide. Field surveys and GNSS/deep displacement monitoring results revealed that the Yanshangou landslide exhibits obvious staged deformation characteristics, and the landslide deformation rate was closely coupled with the dynamic changes in reservoir water level. A slope stability evaluation method integrating the Morgenstern–Price limit equilibrium method and Richard’s equation was established, and the results indicated that the Yanshangou landslide has low saturated permeability. Therefore, its factor of safety (FOS) presents a clear four-stage variation trend in response to reservoir water level fluctuations. A Smoothed Particle Hydrodynamics (SPH)-based numerical model was further developed to simulate the landslide-induced surges under two typical reservoir water level scenarios (815 m and 765 m). The simulation results demonstrated that a high reservoir water level led to more intense surges with greater height and higher velocity, while a low reservoir water level resulted in surges with a wider propagation range along the reservoir bank. The research findings of this study provide a comprehensive theoretical basis and detailed data support for the prevention and mitigation of geological hazards in the Baihetan Reservoir area, and also offer a reference for the hazard management of similar reservoir landslides worldwide.

1. Introduction

As one of the world’s largest hydropower projects, the Baihetan Hydropower Station plays an irreplaceable strategic role in national energy supply, flood control and disaster reduction, and rational allocation of water resources [1,2,3]. However, reservoir impoundment inevitably reshapes the regional hydrological and geomorphological environments, disrupts the original geological stress equilibrium of bank slopes, and easily triggers or exacerbates landslide geological hazards [4,5,6,7]. Reservoir-induced landslides and associated impulse waves have caused catastrophic disasters worldwide and become a classic research topic in engineering geology. The Vajont landslide in Italy is the most representative catastrophic reservoir landslide in history, where massive slope failure generated extremely high impulse waves that overtopped the dam and caused heavy casualties [8,9]. In China, numerous typical reservoir landslide disasters with severe surge hazards have been widely documented, such as the 1985 Xintan landslide and 2003 Qianjiangping landslide in the Three Gorges Reservoir area [10,11], the 2007 Muzhuping landslide in the Shuibuya Reservoir [12], and the 2018 Jinsha River landslide [13] that occurred on the Baihetan Reservoir. These well-documented events demonstrate that large-scale landslides along reservoir banks can induce extremely destructive impulse waves, threatening dam safety and nearby residential areas, as well as navigation. Such reservoir-induced landslides have become a critical constraint on the safe operation of large hydropower reservoirs and the protection of the lives and property of surrounding residents.

The Yanshangou landslide is situated on the northeastern slope of the left bank of the Jinsha River in the Baihetan Reservoir area, approximately 71.2 km from the dam site. It is identified as a high-risk geological hazard point in the region. The landslide is adjacent to the newly built Liugu Town, a key resettlement area for reservoir immigrants with concentrated residential communities and public infrastructure. It therefore poses a possible threat to the safety of the resettlement area. Since the commencement of reservoir impoundment in April 2021, the Yanshangou landslide has exhibited significant deformation signs. Cracks are continuously propagating and expanding, the deformation boundary has become increasingly clear, and local collapses have occurred at the leading edge of the slope. The terrain modification caused by the land leveling project for Liugu Town, the alternating hard and soft lithologic structure, and the intense tectonic activity controlled by the Xiaojiang Fault Zone provided favorable long-term geological and topographic preconditions for the initiation and preparation of the landslide. However, due to the ongoing deformation, the direct impact of reservoir water level fluctuations on the recent deformation, stability evolution, and potential sudden disasters of the downstream slope has become an urgent research priority for the prevention of geological disasters in the Baihetan Reservoir area.

Reservoir-induced landslides are typically characterized by complexity, suddenness, and chain reaction effects, with the interaction between reservoir water level fluctuations and slope geological conditions being the core controlling factor [14,15,16,17,18]. Existing studies have confirmed that reservoir water level changes can alter the pore water pressure distribution in slope rock and soil masses, reduce the shear strength of potential sliding zones, and modify the hydrodynamic conditions of the bank slope, thereby triggering or accelerating slope deformation [19,20,21]. For high-risk landslides adjacent to key infrastructure and residential areas, accurate identification of deformation characteristics, scientific evaluation of stability evolution law under reservoir water level fluctuations, and reliable prediction of secondary hazards (especially landslide-induced surges) are essential prerequisites for formulating targeted and effective disaster prevention and mitigation strategies.

Despite extensive research on reservoir landslides globally, the unique geological background and deformation characteristics of the Yanshangou landslide have not been fully and systematically studied. Current research lacks in-depth exploration of its deformation mechanism, stability evolution law in response to reservoir water level fluctuations, and the hazard intensity of induced surges. To fill this research gap, this study integrates field monitoring, numerical simulation, and theoretical analysis to conduct a comprehensive and systematic investigation of the Yanshangou landslide. The specific research objectives are: (1) to reveal the staged deformation characteristics of the landslide and its evolution law coupled with reservoir water level changes; (2) to establish a reliable slope stability evaluation method and determine the current stability status of the landslide; and (3) to simulate the landslide-induced surges under typical reservoir water level scenarios and quantitatively assess the hazard degree of surges. It should be emphasized that this study only focuses on the effect of reservoir water level fluctuations. Seismic activity and other external triggering factors are not considered in the current research, which will be included in future comprehensive risk assessment. The research findings are expected to provide critical scientific support for the precise prevention and control of the Yanshangou landslide, and also serve as a valuable reference for the hazard management and risk assessment of similar reservoir landslides in other regions.

2. Geological Background

The Yanshangou landslide is a new landslide induced by reservoir impoundment, rather than a reactivated ancient landslide. The construction of the relocated Liugu Town was completed and residents had moved in before the reservoir impoundment. The Baihetan Reservoir started impoundment on 6 April 2021, and significant deformation of the slope occurred immediately after the reservoir water level rose. It can be confirmed that the occurrence and deformation of the landslide were directly triggered by reservoir water level fluctuation and were rarely related to the town construction activities. The following is an overview of the engineering background.

2.1. Topography and Landforms

The Yanshangou landslide zone is located on the northeastern slope of the left bank of the Jinsha River, adjacent to the relocated Liugu Town. It belongs to the geomorphological unit at the eastern foot of the Yingpan Mountain (see Figure 1a). This area is situated within the reservoir zone of the Baihetan Hydropower Station, approximately 71.2 km from the dam site. The Jinsha River flows in a direction of N20°~30°W along this stretch. Due to the commencement of reservoir impoundment in April 2021, the hydrological and geomorphological environment in the surrounding area has undergone significant changes. As a centralized resettlement site for immigrants in the reservoir area, the main construction of the relocated Liugu Town was completed in December 2020, and its leveling project directly reshaped the regional terrain. Before the implementation of the site leveling project for Liugu Town, the original elevation of the top of Yingpan Mountain reached 970 m. After the completion of site leveling, the top elevation was adjusted to approximately 909 m, forming a flat and open platform, becoming an area for immigrants to live and for public facilities layout (see Figure 1b,c).

The terrain presents clear vertical zoning features. The elevation range of 909–875 m is a gentle slope zone, with a gradient of 10–25°, featuring relatively flat terrain. The area below 875 m elevation is a steep slope section where the gradient increases sharply to 30~45°. The slope has low vegetation coverage, dominated by weeds, and a high degree of exposed rock mass. The Yanshan gully develops on the downstream side (left side) of the deformation mass. Originating at an elevation of 909 m, the gully has a cutting depth of 20–30 m and flows in the NNE direction. Affected by the reservoir impoundment, as of 27 May 2022, the section of the gully below 782.13 m elevation has been submerged by reservoir water, forming a special land–water interface geomorphology.

The Yanshangou landslide has a planar shape of an irregular ellipse. The slope length (from rear edge to front edge) is approximately 380 m, and the width is approximately 118 m at an elevation of 880 m and approximately 175 m at the normal reservoir water level of 825 m, with a total deformation area of about 5.45 × 10⁴ m². The sliding depth (thickness) varies from 26.10 m to 48.90 m. The total landslide volume is estimated to be approximately 1.1 × 10⁶ m³. The estimated elevation at the front edge of the notch is approximately 695 m, while the elevation at the rear edge is 903 m. We further clarify that the rear edge of the determined sliding surface is located outside the reconstruction and resettlement scope of Liugu Town, and thus, the deformation and potential failure of the Yanshangou landslide will not affect the reconstructed area.

2.2. Stratigraphy and Lithology

Based on engineering geological mapping, drilling exposure, and regional geological background analysis, the strata in the study area are divided into Silurian System (S), Devonian System (D), and Quaternary System (Q) from oldest to youngest, with a distinct alternating hard and soft lithological combination. The detailed characteristics of each stratum are as follows.

The Silurian System (S) consists of the Middle Series Shimenkan Formation (S₂s) and the Upper Series (S₃). The Middle Series Shimenkan Formation (S₂s) is dominated by hard rocks, primarily gray to dark gray medium-thick to thick-bedded limestone, nodular limestone, and bioclastic limestone. Locally, it is intercalated with grayish-green and dark purple siltstone and mudstone, with a total thickness of more than 90 m and high weathering resistance. This formation is subdivided into two members. The lower member (S₂s¹) is dominated by limestone with a thickness of more than 50 m, and the upper member (S₂s²) is characterized by the alternation of thin-bedded mudstone, siltstone, and thick-bedded limestone with a thickness of 35 to 55 m. Both members are exposed on the eastern hillside of the deformation mass. The Upper Series (S₃) is composed of soft rocks, including purplish-red and dark purple thin to medium-thick bedded mudstone, silty mudstone, and shale, intercalated with blue-gray sandstone and siltstone. It exhibits distinct water sensitivity, softening upon wetting and cracking upon drying, and thus has poor engineering properties. Distributed in a banded pattern at the front edge and on the eastern hillside of the deformation mass with a thickness of 35 to 55 m, this series constitutes a weak layer conducive to landslide development.

The Devonian System (D) consists of the Lower Series (D₁) and the Middle Series Yaopengzi Formation (D₂y). The Lower Series (D₁) is dominated by medium-hard rocks, primarily grayish-white and blue-gray thin to medium-thick-bedded quartz sandstone and siltstone with intercalations of purplish-red and blue-gray mudstone and shale. Its quartz sandstone features hard grains and dense cementation, and thus has high resistance to weathering and erosion. This series outcrops in a banded pattern at the steep scarp of the front edge of the deformation mass, is in conformable contact with the underlying Silurian Upper Series (S₃), has a thickness of approximately 25 to 30 m, and acts as a supporting unit for the front edge of the deformation mass. The Middle Series Yaopengzi Formation (D₂y) represents the main stratigraphic unit of the deformation mass and has a complex lithological assemblage with alternating hard and soft lithologies, including silty mudstone, mudstone, and quartz sandstone, as well as dolomite with siltstone intercalations. With a total thickness of more than 80 m, it is in conformable contact with the underlying Devonian Lower Series (D₁) and is subdivided into three members. The lower member (D₂y¹) is dominated by soft grayish-black silty mudstone and mudstone with a thickness of 25 to 30 m and high water sensitivity. The middle member (D₂y²) is mainly composed of medium-hard quartz sandstone with a thickness of 15 to 20 m and relatively good stability. The upper member (D₂y³) is dominated by hard gray and dark gray dolomite with a thickness of more than 40 m and high weathering resistance. The lower and middle members outcrop at the steep scarp of the deformation mass’s front edge, while the upper member is mainly exposed on both banks of the Yanshangou gully on the left side of the deformation mass.

The Quaternary System (Q) comprises Colluvial–Diluvial Deposits (Q^kel+dl), Alluvial–Proluvial Deposits (Q^apl), and Artificial Accumulations (Q₄^s). The Colluvial–Diluvial Deposits (Q^kel+dl) is yellowish-brown crushed stone mixed soil. Its composition includes boulders (20~40 cm, 15~20% content), crushed stones (6~20 cm, 30~40% content), gravel (0.5~2.0 cm, 25~35% content), and silt. With a loose to slightly dense structure and poor mechanical stability, it is mainly distributed on the surface of the deformation mass above 815 m elevation and the bottom of the Yanshangou gully. Drilling shows its thickness ranges from 2.50 to 4.80 m, and it reaches 10~25 m at the gully mouth. The Alluvial–Proluvial Deposits (Q^apl) is a grayish-yellow pebble mixed soil. It consists of boulders (20~35 cm, 5~10% content), pebbles (6~20 cm, 20~30% content), gravel (0.5~2.0 cm, 20~25% content), silt, and silty clay. Featuring a sub-rounded shape and loose structure, it is only distributed in the Jinsha River floodplain area. Due to periodic submergence by reservoir water, the stratum maintains a high saturation level for a long time. The Artificial Accumulations (Q₄^s) is a yellowish-brown artificial fill, mainly composed of crushed stone mixed soil. Its parent rocks are mainly dolomite, quartz sandstone, and siltstone, with a loose and uneven structure. Formed during the construction of the Liugu relocated market town (main construction completed in 2021), it is distributed on the slope outside the retaining wall of the health center with a thickness of approximately 2~5 m.

The overall characteristics of the shallow earth in the landslide area can be summarized as follows. The deformed rock mass and intensely weathered bedrock are characterized by a rock quality designation (RQD) of 0–8%. The rock mass is broken to relatively broken. The surface layer of the landslide is covered by gravelly mixed soil, and the lower part consists of highly weathered, fragmented hard-soft alternating strata. Under long-term tectonic disturbance and reservoir wetting–drying cycles, the entire sliding mass has gradually evolved into a soil-like mass with discontinuous, low-permeability, and shear-controlled mechanical behavior, which provides the basis for the subsequent hydrogeological and stability analysis.

2.3. Geological Structure

Based on the regional tectonic background and survey results, the area where the deformation mass is located has intense tectonic activity, and the integrity of the rock mass is significantly affected by tectonic actions. The specific characteristics are as follows:

As shown in Figure 2, the deformation mass is situated at the tectonic junction zone between the western part of the Yangtze Paraplatform, the Upper Yangtze Depression, and the Kangdian Axis, directly influenced by the regional active fault, namely the Xiaojiang Fault Zone (F₁₂). As the boundary fault between the Kangdian Axis and the Upper Yangtze Depression, this fault is the core strong seismic activity zone in eastern Yunnan, with a total length of approximately 400 km. It has a complex internal structure, composed of multiple secondary shear faults and tension-shear faults, and has been in an active state for a long time. The northern section of the Xiaojiang Fault passes through the north–south groove of the former Liugu Township, about 750 m west of the deformation mass, with an overall strike of N5°W. Its long-term tectonic movement not only damages the integrity of the regional rock mass but also forms a widely developed joint and fracture system, resulting in a fragmented state of the rock mass in the engineering area and providing favorable tectonic conditions for groundwater seepage and rock mass deformation. During this survey, a reverse fault f1 was successfully exposed at the depth of 94.00~95.00 m in borehole LGK01, and its attitude is inferred to be N42°W, NE < 66°. The rocks within the fault zone have undergone schistosity transformation due to intense tectonic action, forming typical schistose tectonites. The core is broken into gravel-like fragments, with significantly reduced mechanical strength, becoming a weak structural plane in the regional rock mass, which may control the sliding boundary of the deformation mass.

The attitude of the rock strata in the deformation mass and its surrounding area shows regular changes under the influence of tectonic movement. On the eastern side of the deformation mass, from south to north, the rock stratum attitude gradually transitions from N15°~20°E, NW < 40~45° to N30°~40°E, NW < 40~50°. The gradual change in attitude reflects the directional influence of the regional tectonic stress field. Inside and on the western side of the deformation mass, the rock stratum attitude is relatively stable, mainly N30°~40°E, NW < 40~55°, forming an oblique-transverse medium-dipping slope structure with the bank slope direction. This structural combination is prone to bedding-parallel deformation under the action of reservoir water.

3. Materials and Methods

The Yanshangou landslide mass is a soil-like mass composed of surface gravelly mixed soil and underlying highly weathered, fragmented rock mass. Although the parent materials are sedimentary rocks, long-term deformation, weathering, and reservoir water interaction have caused the sliding mass to lose intact rock structure and exhibit soil-like hydrogeological and mechanical properties. Therefore, unsaturated soil mechanics and limit equilibrium methods are used to characterize its seepage and stability evolution.

3.1. Deformation Observation

To systematically clarify the surface and deep deformation laws of the Yanshangou landslide, characterize its deformation features, and accurately evaluate the evolutionary trend of deformation, this study conducted comprehensive field investigations and continuous on-site monitoring, with the monitoring period spanning from 14 July to 13 October 2022; for surface deformation monitoring, GNSS technology was adopted, and four monitoring points were deployed (two inside and two outside the deformation mass), distributed along the main sliding direction and key external positions to achieve full coverage (Figure 2). The monitoring frequency was dynamically adjusted according to the deformation status. Multi-dimensional displacement data of each GNSS monitoring point were collected to calculate composite displacement and displacement rate, followed by correlation analysis with reservoir water level changes. For deep deformation monitoring, inclinometer casings were installed in existing boreholes LGK01~LGK03 along Profile 1, forming three monitoring points that fully penetrate potential slip zones (Figure 3). High-precision inclinometers with a measurement accuracy of ±0.02 mm/m were used for data collection at a frequency of once a week. The depth range of slip zones was identified by analyzing the abrupt change characteristics of displacement data at different depths, so as to track the development and evolutionary process of internal deformation within the landslide mass.

3.2. Slope Stability Evaluation

Slope stability evaluation is accomplished via the integrated application of the Morgenstern–Price limit equilibrium method and Richard’s equation [22,23]. Initially, Richard’s equation is employed to characterize the evolutionary process of the seepage field within unsaturated slopes, enabling the acquisition of spatial distribution features of transient water content and matric suction in the slope. Subsequently, the aforementioned seepage calculation results are coupled with the Morgenstern–Price limit equilibrium method to establish a dynamic correlation between the shear strength of unsaturated soil and the seepage state, thereby completing the quantitative evaluation of slope stability [24,25].

3.2.1. Richard’s Equation for Seepage Field Calculation

Richard’s equation is a partial differential equation describing water movement in unsaturated soil, derived from the combination of Darcy’s law and the law of mass conservation. Its specific expression is as follows [26]:

$${\displaystyle \frac{\mathsf{\partial }\theta }{\mathsf{\partial }t}}=\nabla ·\left[K\left(\psi \right)\nabla \left(\psi +z\right)\right]$$

(1)

where $\theta$ denotes volumetric water content; $t$ represents time; $\psi$ is pore water pressure head, which is a negative value equivalent to the matric suction head; $z$ stands for the elevation head, also known as the gravitational head; and $K\left(\psi \right)$ is the unsaturated hydraulic conductivity function, which is a function of matric suction.

The left-hand side of the equation signifies the time rate of change in water content at a specific point, while the right-hand side represents the net water inflow. It describes the process of water flow in soil driven by the combined effects of gravity and matric suction gradient, resulting in temporal variations in water content at various points. Solving Richard’s equation necessitates two key constitutive relationships: (i) soil-water characteristic curve (SWCC), which depicts the corresponding relationship between matric suction and water content or degree of saturation, intuitively reflecting the soil’s water-holding capacity; (ii) hydraulic conductivity function (HCF), which characterizes the correlation between unsaturated hydraulic conductivity and matric suction or water content.

3.2.2. Determination of SWCC and HCF

As the core foundation for unsaturated soil mechanics analysis, the SWCC model proposed by Fredlund et al. (1994) covers the entire range of suction (0–10⁶ kPa) [24,27]. Its core equation is as follows:

$${\theta }_v=C\left(h\right){\displaystyle \frac{{\theta }_s}{{\left[\mathrm{I}\mathrm{n}{\left(e+\frac{h}{a_f}\right)}^{b_f}\right]}^{c_f}}}$$

(2)

$$C\left(h\right)=1-{\displaystyle \frac{In\left(1+\frac{h}{h_r}\right)}{In\left(1+\frac{{10}^6}{h_r}\right)}}$$

(3)

where θ_v is the volumetric water content of the soil; h is the suction force of the substrate (unit: kPa); θ_s is the saturated volumetric water content (equivalent to soil porosity); e is the natural constant (approximately 2.718); h is matric suction; C(h) is the correction factor ensuring water content approaches 0 when suction reaches 10⁶ kPa; a_f, b_f, c_f are fitting parameters related to soil air entry value, curve inflection slope, and residual water content, respectively; and h_r is a constant associated with matrix suction corresponding to residual water content.

These SWCC model parameters are primarily determined through two methods: experimental fitting, where data on moisture content at different suction levels are measured and fitted using nonlinear least squares, and estimation based on soil properties. According to Houston et al. (2006), non-plastic soils exhibit no significant plastic characteristics, and their SWCC is primarily governed by particle size distribution (PSD) because pore size is directly related to particle grading [28]. The core procedure for deriving parameters based on the PSD curve is elaborated as follows. Firstly, the original particle size data of the Yanshangou landslide are collected, as shown in

Table 1. Particle size composition data of the Yanshangou landslide mass obtained from grain size analysis.

Particle Size (mm)	0.005	0.075	0.25	0.5	2	5	20	60	200
Cumulative percentage of particles smaller than corresponding size (%)	2.5	7.8	9.0	9.5	10.8	12.7	28.5	51.0	100

, and key particle size parameters are extracted from the PSD curve.

Subsequently, the corresponding derived particle sizes are calculated based on these key parameters. Specifically, as shown in Figure 4 and

Table 2. Partial particle size parameters of the Yanshangou landslide mass.

Particle Size Parameter	D0	D10	D20	D30	D60	D90	D100	P200
Unit	mm	mm	mm	mm	mm	mm	mm	%
Value	0.2	1.1	11.9	22.7	85.7	171.4	216.0	7.8

, core parameters, including D₁₀, D₂₀, D₃₀, D₆₀, D₉₀ (corresponding to particle sizes at the cumulative passing rates of 10%, 30%, 60%, and 90%, respectively), and P₂₀₀ (mass fraction of particles passing the No. 200 sieve, i.e., particle size ≤ 0.075 mm) were obtained.

Additionally, derived parameters such as D₁₀₀ (particle size at the 100% cumulative passing rate) and D₀ (pore characteristic particle size) are calculated using the following formulas:

$$\begin{array}{c}{D}_{100}={10}^{{\displaystyle \frac{4\left(\mathit{log}{D}_{90}-\mathit{log}{D}_{60}\right)}{3}}+\mathit{log}{D}_{60}}\end{array}$$

(4)

$$D_0={10}^{{\displaystyle \frac{-3\left(\mathit{log}{D}_{30}-\mathit{log}{D}_{10}\right)}{2}}+\mathit{log}{D}_{30}}$$

(5)

Subsequently, the parameters of the SWCC model are derived using empirical equations as follows:

$$\{\begin{array}{c}{a}_f=1.14a-0.5\\ a=-2.79-14.1\mathit{log}{D}_{20}-1.9\times {10}^{-6}{P}_{200}^{4.347}\mathit{log}{D}_{30}+0.055D_{100}\\ b_f=0.936b-3.8\\ b=\left(5.39-0.29\mathit{ln}\left(P_{200}\cdot {\displaystyle \frac{D_{90}}{D_{10}}}\right)+3{D_0}^{0.57}+0.021P_{200}^{1.19}\right)\cdot {\left({\displaystyle \frac{30}{\mathit{log}{D}_{90}-\mathit{log}{D}_{60}}}\right)}^{0.1}\\ c_f=0.26e^{0.758c}+1.4D_{10}\\ c=\mathit{log}{\left({\displaystyle \frac{20}{\mathit{log}{D}_{30}-\mathit{log}{D}_{10}}}\right)}^{1.15}-1+{\displaystyle \frac{1}{b_f}}\\ h_r=100 kPa\end{array}$$

(6)

Finally, these parameters were substituted into Fredlund’s SWCC equation to obtain the complete suction–water content relationship curve, as shown in Figure 5.

Second, Fredlund et al. (1994) derived an integral form of the HCF based on the SWCC model [27]. This is because soil permeability is fundamentally controlled by the pore size distribution, and since the SWCC model already indirectly reflects the characteristics of this distribution, the HCF can be predicted through integral operations on the SWCC model. To eliminate variations in inherent permeability across different soils and enhance the universality of the HCF, the relative permeability k_r(ψ) is introduced, defined as:

$$k_r\left(\psi \right)={\displaystyle \frac{k\left(\psi \right)}{k_s}}$$

(7)

where k(ψ) is the unsaturated hydraulic conductivity of a given matrix at suction ψ; k_s is the saturated hydraulic conductivity of the soil.

Based on Fredlund’s SWCC model, an integral expression for relative permeability is derived through integration along the suction axis, fully accounting for the nonlinear relationship between water content and substrate suction.

$$k_r\left(\psi \right)={\displaystyle \frac{{\int }_{\frac{In\left({\psi }_{aev}\right)}{In\left({10}^6\right)}}^1{e}^y\frac{\theta \left(e^y\right)-{\theta }_s}{{\theta }_s-\theta \left(e^y\right)}{\theta }^{\prime }\left(e^y\right)dy}{{\int }_{\frac{In\left(\psi \right)}{In\left({10}^6\right)}}^1{e}^y\frac{\theta \left(e^y\right)-\theta \left(\psi \right)}{{\theta }_s-\theta \left(e^y\right)}{\theta }^{\prime }\left(e^y\right)dy}}$$

(8)

where ψ_aev is the soil air entry value, representing the critical matric potential at which unsaturated soil begins drainage; θ′(e^y) is the first derivative of suction in the SWCC equation, reflecting the rate of change in water content with matrix suction and characterizing the steepness or gentleness of pore distribution; and θ(e^y) is the volumetric water content corresponding to suction e^y.

However, pore tortuosity reduces the effective cross-sectional area of the actual permeation path, thereby decreasing the hydraulic conductivity. To account for the influence of soil pore tortuosity on permeability characteristics, a power-law correction term Θ^q based on normalized moisture content (relative saturation) can be introduced to further enhance the predictive accuracy of HCF across different soil types. A power-law exponent q = 1 is recommended. The relative permeability expression after introducing the correction term Θ^q(ψ) is:

$$k_r\left(\psi \right)={\Theta }^q\left(\psi \right)·{\displaystyle \frac{{\int }_{\frac{In\left({\psi }_{aev}\right)}{In\left({10}^6\right)}}^1{e}^y\frac{\theta \left(e^y\right)-{\theta }_s}{{\theta }_s-\theta \left(e^y\right)}{\theta }^{\prime }\left(e^y\right)dy}{{\int }_{\frac{In\left(\psi \right)}{In\left({10}^6\right)}}^1{e}^y\frac{\theta \left(e^y\right)-\theta \left(\psi \right)}{{\theta }_s-\theta \left(e^y\right)}{\theta }^{\prime }\left(e^y\right)dy}}$$

(9)

Therefore, using the landslide mass at Yanshangou as the study subject, the relationship curve between matrix suction and relative permeability coefficient was obtained through SWCC data (see Figure 6).

3.2.3. Shear Strength Criterion of Unsaturated Soil

To apply the results of seepage analysis to stability calculations, a shear strength formula considering the contribution of matric suction is required. The most well-known is the extended Mohr–Coulomb criterion proposed by Fredlund et al. (1993) [24].

$${\tau }_f=c^{\prime }+\left({\sigma }_n-u_a\right)\mathit{tan}{\varphi }^{\prime }+\left(u_a-u_w\right)\mathit{tan}{\varphi }^b$$

(10)

where ${\tau }_f$ is shear strength; $c^{\prime }$ is effective cohesion; ${\sigma }_n$ is total normal stress; $u_a$ is pore air pressure, which is usually assumed to be atmospheric pressure, i.e., $u_a$ = 0; $u_w$ is pore water pressure (negative value); ${\varphi }^{\prime }$ is effective internal friction angle; and $u_a$ − $u_w$ is matric suction.

In practice, for simplification, it is often assumed that the ratio of $\tan {\varphi }^b$ to $\tan {\varphi }^{\prime }$ is a constant $\chi$ (generally ranging from 0 to 1), or more simply, ${\varphi }^b$ ≈ ${\varphi }^{\prime }$ directly. Thus, the formula can be simplified as follows [29].

$${\tau }_f=c^{\prime }+\left({\sigma }_n-u_a\right)\mathit{tan}{\varphi }^{\prime }+\left(u_a-u_w\right)\mathit{tan}{\varphi }^{\prime }$$

(11)

This formula quantifies the contribution of negative pore water pressure (matric suction) to shear strength in the unsaturated zone. The matric suction can be determined from the seepage field calculation results obtained by solving Richard’s equation, so the shear strength at each point of the slope can be directly updated using transient matric suction.

3.2.4. Morgenstern–Price Limit Equilibrium Method

The Morgenstern–Price method, a type of limit equilibrium method, is adopted for slope stability analysis in this study. Its core idea is to vertically divide the potential sliding mass into several slices. Subsequently, the forces and moments acting on each slice are analyzed, and the safety factor of the entire sliding mass is solved by simultaneously solving the static equilibrium equations and introducing an inter-slice force function, as shown in the following formula [22].

$$F_s={\displaystyle \frac{\sum \left[c^{\prime }{L}_i+\left(W_i\mathit{cos}{\alpha }_i-U_i+E_i\mathit{sin}{\alpha }_i-X_i\mathit{cos}{\alpha }_i\right)\mathit{tan}{\varphi }^{\prime }\right]}{\sum \left[W_i\mathit{sin}{\alpha }_i+E_i\mathit{cos}{\alpha }_i+X_i\mathit{sin}{\alpha }_i\right]}}$$

(12)

where $W_i$ is self-weight of the i-th slice; ${\alpha }_i$ is inclination angle of the sliding surface of the i-th slice; $U_i$ is pore water pressure on the sliding surface of the i-th slice; $E_i$ and $X_i$ are inter-slice normal force and tangential force of the i-th slice, respectively; and $L_i$ is length of the sliding surface of the i-th slice.

The geomechanical parameters used in the stability analysis were determined based on laboratory tests and engineering geological characteristics of the strata. The unit weight, cohesion, and internal friction angle are 21 kN/m³, 10 kPa, and 30°, respectively.

3.3. Pre-Analysis of Landslide-Induced Surge Hazards

The process of landslide-induced impulse waves involves violent free-surface evolution and complex fluid–structure interaction effects. Traditional Eulerian grid-based numerical methods often encounter mesh distortion and entanglement issues when tracking free interfaces, which limit their effectiveness in simulating such highly nonlinear processes and constrain their application in landslide tsunami research. In contrast, Smoothed Particle Hydrodynamics (SPH), as a mesh-free Lagrangian discretization method, characterizes the evolution of fluid parcels through particle motion, offering distinct advantages in handling large-deformation flows and free-surface fragmentation [30,31]. Consequently, this study develops an SPH-based numerical framework to accurately characterize the complete disaster chain—from landslide initiation and surge generation to wave.

3.3.1. SPH Basic Control Equations

As a mesh-free Lagrangian discretization approach, the SPH method represents the continuous computational domain through a finite set of discrete particles. Within this framework, each particle serves both as a carrier of physical properties and as a computational node for field function interpolation, and the trajectories of these particles are governed by the Lagrangian form of the Navier–Stokes equations [32]. Under the Weakly Compressible SPH (WCSPH) framework, the governing equations for mass and momentum conservation can be formulated as follows:

$${\displaystyle \frac{d{\rho }_i}{dt}}={\rho }_i\sum _j{\displaystyle \frac{m_j}{{\rho }_j}}{\nu }_{ij}\cdot {\nabla }_i{W}_{ij}$$

(13)

$${\displaystyle \frac{d{\mathit{v}}_i}{dt}}=-\sum _j{m}_j\left({\displaystyle \frac{P_i}{{{\rho }_i}^2}}+{\displaystyle \frac{P_j}{{{\rho }_j}^2}}+{\Pi }_{ij}\right){\nabla }_i{W}_{ij}+\mathit{g}$$

(14)

where subscript i denotes the interpolated particle and subscripts, j denotes the neighboring particle of particle i, $\rho$ and m denote the density and mass of the fluid particle, respectively, P is the pressure, $\nu$ represents the fluid velocity, and $\mathit{g}$ is the gravitational acceleration. ${\nu }_{ij}$ signifies the velocity vector difference between particle i and particle j, and ${\nabla }_i{W}_{ij}$ denotes the gradient of the kernel function $W$. ${\Pi }_{ij}$ is the artificial viscosity used to suppress numerical oscillations, calculated as [33]:

$${\Pi }_{ij}=\{\begin{array}{cc}-{\displaystyle \frac{\alpha C_{ij}{\mu }_{ij}}{{\rho }_{ij}}},& v_{ij}\cdot r_{ij}<0\\ 0,& v_{ij}\cdot r_{ij}\ge 0\end{array}$$

(15)

where $\alpha$ represents the artificial viscosity coefficient, h is the smoothing length of kernel function $W$, and $C_{ij}$ and ${\rho }_{ij}$ are the average speed of sound and density between particles i and j, respectively. The term ${\mu }_{ij}$ is defined as:

$${\mu }_{ij}={\displaystyle \frac{h{v}_{ij}·r_{ij}}{r_{ij}^2+0.01h^2}}$$

(16)

where $r_{ij}$ is the inter-particle distance, and $0.01h^2$ is a regularization parameter to prevent singularities. Following the study by Shao and Lo (2003), the artificial viscosity coefficient $\alpha$ is typically constrained within the range of 0.01 to 0.1 to balance numerical robustness with physical fidelity [34].

In the WCSPH method, the governing equations are closed by an equation of state (EOS) that defines the relationship between pressure and density. The EOS enables the direct calculation of pressure gradients between particles, thereby simplifying the treatment of complex fluid dynamics equations. In this study, the widely used Tait equation of state is employed to maintain weak compressibility by utilizing a numerical speed of sound that is lower than the physical speed of sound [35]. The Tait equation is expressed as follows:

$$P=B\left[{\left({\displaystyle \frac{\rho }{{\rho }_0}}\right)}^{\gamma }-1\right]$$

(17)

where $\gamma =7$; ${\rho }_0$ is a reference density, ${\rho }_0=1000 \mathrm{kg}/{\mathrm{m}}^3$; and B is the pressure constant, $B=C_0^2{\rho }_0/\gamma$, where $c_0$ is the speed of sound at the reference density. Given that the fluid is treated as weakly compressible, the numerical speed of sound is constrained to be at least ten times the maximum anticipated fluid velocity. This constraint ensures that density fluctuations remain sufficiently small (typically below 1%), thereby maintaining quasi-volume conservation. Furthermore, the fluid pressure can be explicitly solved through the aforementioned equation of state, facilitating a numerical implementation that effectively leverages parallel computing architectures.

3.3.2. Non-Newtonian Rheological Model for Landslides

Existing research indicates that the relationship between the shear strain rate and shear stress of highly deformable soils closely approximates that of a Bingham fluid; thus, non-Newtonian fluid models can be employed to describe the dynamic characteristics of landslides [36,37,38]. To describe the dynamic characteristics of landslides, the generalized Herschel–Bulkley–Papanastasiou (HBP) model is adopted, which is capable of representing a wide range of viscoplastic behaviors [39]. The constitutive relationship for the HBP model is defined by the shear stress tensor $\tau$ as:

$$\tau =2\eta D$$

(18)

where $D$ is the symmetric shear rate tensor (the rate-of-deformation tensor). The fundamental component of this model is the apparent viscosity $\eta$, formulated as:

$$\eta ={\displaystyle \frac{{\tau }_y[1-\mathit{exp}(-m|\dot{\gamma }\left|\right)]}{\left|\dot{\gamma }\right|}}+K|\dot{\gamma }{|}^{n-1}$$

(19)

where ${\tau }_y$ denotes the yield stress, K represents the consistency index, n is the power-law index (characterizing shear-thinning for n < 1), and $\dot{\gamma }$ is the symmetric strain rate tensor. The model incorporates the Papanastasiou parameter m, which controls the exponential growth of stress at low shear rates. Unlike the original Herschel–Bulkley model, the HBP formulation effectively resolves the numerical discontinuity at zero shear rates by ensuring that the stress reaches a finite value in the unyielded region, thereby significantly enhancing numerical stability.

By substituting the apparent viscosity $\eta$ calculated from the HBP model into the momentum equation, the governing equations for the landslide can be reformulated as follows:

$${\displaystyle \frac{d{\mathit{v}}_i}{dt}}=-\sum _j{m}_j\left({\displaystyle \frac{P_i}{{\rho }_i^2}}+{\displaystyle \frac{P_j}{{\rho }_j^2}}\right){\nabla }_i{W}_{ij}+\sum _j{m}_j{\displaystyle \frac{({\eta }_i+{\eta }_j)r_{ij}\cdot {\nabla }_i{W}_{ij}}{\left({\rho }_i{\rho }_j\right)(r_{ij}^2+0.01h^2)}}{\mathit{v}}_{ij}+\mathit{g}$$

(20)

where $\eta$ represents the apparent viscosity, which is dynamically updated for landslide particles based on the HBP model while remaining constant for water particles; $r_{ij}$ is the corresponding inter-particle distance. By employing this unified variable-viscosity approach, the momentum exchange between the landslide and the water is naturally captured within the viscous term, enabling a high-fidelity representation of the intense fluid–structure interaction during surge generation.

In this paper, the SPH-based numerical framework is employed to simulate the Yanshangou landslide in the Baihetan Reservoir area. Based on engineering geological survey data, the reservoir water level in the downstream is maintained at 765 m, and the computational domain is discretized using an initial particle spacing of d_p = 4 m. The resulting total number of particles for the landslide and the water body are 2.59 × 10⁵ and 4.19 × 10⁶, respectively. The dynamic characteristics of the landslide are characterized by the HBP model. To approximate the rheological behavior of the landslide material as a Bingham fluid, the power-law index n is set to 1. Furthermore, the Papanastasiou regularization parameter m is set to 100 to smooth the stress discontinuities at low shear rates, thereby ensuring the numerical robustness of the simulation process. Detailed physical and numerical parameters are summarized in

Table 3. Physical and numerical parameters used in the SPH simulation.

Particle Size Parameter	Unit	Selected Value
CFL number	/	0.2
Coefficient of speed of sound	/	10
Polytropic index	/	7
Number of steps to apply Euler time stepping	/	40
Particle distance	m	4
viscosity	m2/s	0.1
Density of water	kg/m3	1000
Density of landslide	kg/m3	1650
Interaction kernel function	/	Wendland
Time-stepping algorithm	/	Velocity-Verlet
Viscosity treatment	/	Artificial
Boundary treatment	/	Dynamic boundary condition

.

4. Results

Initial deformation signs of the Yanshougou landslide first emerged in late November 2021. It entered an accelerated deformation phase in May 2022. Affected by reservoir impoundment with the reservoir water level at approximately 781.11 m at that time, slope cracks developed and propagated rapidly, the deformation boundary of the bank slope became distinct, and the overall shape presented an irregular arc. Collapse occurred at the front edge of the reservoir-adjacent bank slope, with the maximum elevation of the collapsed bank reaching about 840 m. As the reservoir water level rises, the scope of bank collapse will further expand. By 5 August 2022, the reservoir water level dropped to 775.89 m, yet deformation continued to progress. Meanwhile, the number and scale of cracks increased. From then until October, the deformation rate gradually decreased, maintaining a slow deformation state.

4.1. Deformation Characteristics

Two on-site surveys were conducted to investigate the deformation characteristics and crack development of the deformable mass in the study area. The distribution and development characteristics of these cracks are clearly illustrated in Figure 7. The first survey was carried out on 27 May 2022, during which a total of 17 large-scale cracks were identified, with tensile cracking and downthrow as the main deformation features. The core crack L2 distributes in an arc shape, with an extension length of 159 m, a maximum width of 40 cm, a downthrow height of approximately 80 cm, and a visible depth of 4.2 m. It penetrates from the rear edge at an elevation of 902 m to the right side and pinches out at an elevation of 854 m. Three cracks (L10, L11, and L12) on the left side develop in a plumose pattern, with lengths ranging from 10 to 40 m and widths from 20 to 30 cm, and local compressive uplift is observed in some areas. On the right side, cracks L3 and L4 have lengths of 17.8 m and 32 m respectively, widths of 15 to 20 cm, and extend along the slope to an elevation of 826 m, where they pinch out. Internal transverse cracks in the study area are concentrated at elevations of 870~900 m, while longitudinal cracks are distributed at elevations of 865~825 m. Both types of cracks mostly have lengths of 10~40 m and widths of 5~30 cm.

The second survey was performed on 5 August 2022, and the crack development showed significant characteristics of increased quantity and expanded scale. A total of 29 cracks were detected in this survey. The extension length of the core crack L2 increased to 161 m, the maximum width expanded to 150 cm, and the downthrow height reached 85 cm, which has completely penetrated the rear edge and both side boundaries. Three new plumose cracks (L25, L26, and L27) were added on the left side, with the longest one being 31.7 m in length and 3~90 cm in width. The downthrow amount on the river-adjacent side is approximately 30 cm, and it extends to an elevation of 819 m. The length of crack L4 on the right side increased to 48.8 m, with a width of 18 cm, extending to an elevation of 818 m. Both the length and width of internal transverse and longitudinal cracks have expanded to varying degrees, and some cracks that originally exhibited compressive deformation characteristics have transformed into tensile deformation.

Both monitoring points TPlg02 and TPlg03 within the landslide mass exhibit significant deformation response characteristics. Specifically, the cumulative 3D resultant displacement of TPlg02 reaches 210.21 mm with a displacement direction of N25°E, while that of TPlg03 amounts to 240.48 mm oriented at N11°E. The displacement directions of both points are completely consistent with the main sliding direction of the landslide, which directly verifies the deformation trend of the landslide along the detected sliding surface. As shown in Figure 8, the deformation rates of the two monitoring points show significant dynamic fluctuation characteristics over time, and they have a clear coupling relationship with the dynamic changes in the reservoir water level. At the initial stage of monitoring (i.e., from 14 July to 29 July 2022), the reservoir water level dropped continuously from 784.6 m to the lowest value of 776.4 m during the monitoring period, and the deformation rates rose sharply to reach the peak values of the monitoring period. During this period, the maximum deformation rate of TPlg02 was 6.0 mm/d, and that of TPlg03 was 7.6 mm/d. After 9 August 2022, the reservoir water level entered a gradual recovery phase, and the deformation rates decreased continuously accordingly. After 19 August, the reservoir water level dropped slightly again, and the deformation rates showed a synchronous upward trend, which further verified the controlling effect of reservoir water level changes on the deformation rates. During the period from 29 August to 13 September 2022, the reservoir water level maintained a stable state, and the deformation rates of TPlg02 and TPlg03 stabilized correspondingly at 2.0 mm/d and 2.1 mm/d, respectively. After this stage, the reservoir water level continued to rise, and the deformation rates showed a gradual attenuation trend once again. In sharp contrast to the monitoring points inside the landslide mass, the external reference points TPlg01 and TPlg04 outside the landslide mass only exhibit minor deformation characteristics, with deformation magnitudes far lower than those of the internal points. Among them, the cumulative 3D resultant displacement of TPlg01 is only 4.97 mm, and that of TPlg04 is 10.87 mm. This indicates that the deformation of the landslide mass has not extended to the external areas, and the surrounding bank slopes are in a stable state as a whole.

During the monitoring period, the displacement data of the inclinometer hole INlg01, which is located outside the deformation mass, showed no abnormal fluctuations up to 3 October 2022. This indicates that the deep rock mass in this area did not participate in deformation activities and remained in an overall stable state. In contrast, inclinometer holes INlg02 and INlg03, both situated within the deformation mass, exhibited distinct depth-stratified deformation characteristics, with deformation concentrated in the shallow to mid-deep zones. Specifically, the deformation of the inclinometer hole INlg02 showed an obvious, abrupt variation. The rock mass above the depth of 26.50 m underwent significant displacement toward the main sliding direction of the landslide, with the cumulative maximum displacement reaching 69.38 mm. In the depth interval of 25.50–27.50 m, the displacement attenuated sharply, forming a clear deformation boundary (see Figure 9a). This abrupt deformation zone corresponds to the upper member of the Middle Devonian Yaopengzi Formation (D₂y³), with dolomite as the lithology. Field-exposed rock cores were in a gravelly, fragmented state, and clear compression and folding traces were observed in the 25.50–26.10 m interval. Combined with the characteristics of abrupt displacement, this interval is determined to be the main slip zone of the landslide. For the inclinometer hole INlg03, deformation was concentrated in the zone above the depth of 50 m. The rock mass within this depth range continuously displaced toward the main sliding direction, with the cumulative maximum displacement reaching 128.35 mm. After the depth of 45 m, the displacement showed a rapid decreasing trend; beyond 50 m, the displacement of the rock mass was nearly zero, showing almost no deformation response (see Figure 9b). The abrupt deformation zone of this hole corresponds to the lower member of the Middle Devonian Yaopengzi Formation (D₂y¹), with gray-black mudstone as the lithology. Field rock cores were fragmented, and continuous compression and slip surfaces developed in the 48.0–48.9 m interval. Based on a comprehensive analysis of deformation characteristics and microscopic rock core evidence, this interval is inferred to be the slip zone corresponding to the inclinometer hole INlg03.

4.2. FOS Calculation Result

Given that the actual saturated permeability coefficient of the Yanshangou landslide has not yet been determined through field tests, this study employs numerical simulations with multiple permeability coefficient scenarios to inversely infer its actual permeability characteristics. To accurately determine the stability evolution law of the Yanshangou landslide and clarify the influence mechanism of saturated permeability coefficient on landslide stability, numerical simulations were carried out under five different saturated permeability coefficient scenarios (1 × 10⁻⁷ m/s, 1 × 10⁻⁶ m/s, 1 × 10⁻⁵ m/s, 1 × 10⁻⁴ m/s, and 1 × 10⁻³ m/s). The FOS of the landslide was calculated, and the dynamic correlation between the simulated FOS and the field-measured actual deformation rate was systematically analyzed to verify the rationality of the simulation results and reveal the intrinsic link between permeability characteristics and landslide stability.

The saturated permeability coefficient exhibits a significant regulatory effect on the temporal evolution of the landslide’s FOS. Under the low saturated permeability coefficient scenario (1 × 10⁻⁷ m/s, Figure 10a), the FOS of the landslide presents a distinct four-stage variation trend. Moreover, this trend is highly coupled with the fluctuation of the reservoir water level and the change in deformation rate. At stage 1 (14 July–1 August 2022), the reservoir water level declined rapidly from 784.6 m to 776.4 m. Given the low permeability of the landslide mass, the pore water pressure inside could not be dissipated timely and effectively (Figure 11), which generated significant hydrodynamic pressure along the potential sliding surface. The increased seepage water pressure directly raised the downslope driving force acting on the landslide mass, leading to a sharp decline in the factor of safety (FOS) from 1.16 to 1.10. Simultaneously, the landslide deformation rate increased drastically, with the maximum deformation rate at monitoring point TPlg03 reaching 7.6 mm/d, which is indicative of an accelerated deformation state of the landslide. At stage 2 (1 August–19 August 2022), the reservoir water level entered a phase of gradual recovery. As the reservoir water level rose, the pore water pressure inside the landslide underwent a gradual adjustment, leading to a partial restoration of the shear strength of the sliding mass. The FOS exhibited a slow upward trend, increasing from 1.10 to 1.15. Correspondingly, the landslide deformation rate continued to decline, which reflects a gradual improvement in the stability of the slope. At stage 3 (19 August–11 September 2022), a slight secondary decline in the reservoir water level occurred during this period. Given the low permeability characteristic, the landslide presented a high sensitivity to variations in the reservoir water level. The minor drop in reservoir water level triggered a redistribution of pore water pressure within the landslide, and the FOS decreased synchronously to 1.11. The landslide deformation rate also showed a slight upward tendency, which further verifies the close response correlation between the FOS and reservoir water level changes under low permeability conditions. At stage 4 (11 September–13 October 2022), the reservoir water level sustained a rising trend. Due to the low permeability coefficient of the landslide mass, the groundwater level inside the landslide could not rise in synchronization with the reservoir water level, thus resulting in the formation of hydrodynamic pressure with a direction favorable for shear resistance. This reverse osmotic pressure exerted a positive effect on enhancing slope stability, causing the FOS to increase gradually to 1.48. The landslide deformation rate displayed a progressive attenuation trend, and the slope tended to a stable state overall.

As depicted in Figure 10b, the overall variation trend of the FOS is consistent with that of the 1 × 10⁻⁷ m/s scenario. Specifically, it exhibits a four-stage variation pattern coupled with reservoir water level fluctuation and deformation rate. However, the magnitude of FOS variation is smaller. For the scenarios with saturated permeability coefficients of 1 × 10⁻⁵ m/s, 1 × 10⁻⁴ m/s, and 1 × 10⁻³ m/s (as shown in Figure 10c–e), the temporal evolution patterns of the FOS are significantly inconsistent with the actual deformation behaviors and stability responses of the Yanshangou landslide obtained from field observations and monitoring.

By comparing the FOS simulation results across different saturated permeability coefficient scenarios with the field-measured deformation rates, it is confirmed that the simulation outcomes under low saturated permeability coefficient conditions (1 × 10⁻⁷ m/s and 1 × 10⁻⁶ m/s) are in good agreement with the actual behaviors of the Yanshangou landslide. Under these scenarios, the FOS exhibits a sensitive response to fluctuations in the reservoir water level, and its temporal evolution trend is highly coupled with the deformation rate of the landslide. Specifically, during the stage of rapid reservoir water level drawdown, the FOS decreases significantly, while the deformation rate surges; in the subsequent stages of reservoir water level recovery and stabilization, the FOS stabilizes with a slight upward tendency, and the deformation rate attenuates accordingly. This variation pattern is fully consistent with the field monitoring data, accurately reflecting the actual stability response of the landslide to reservoir water level changes. In contrast, the simulation scenarios with saturated permeability coefficients of 1 × 10⁻⁵ m/s, 1 × 10⁻⁴ m/s, and 1 × 10⁻³ m/s overestimate the pore water pressure dissipation capacity of the Yanshangou landslide. As a result, the predicted FOS evolution trends and deformation rates deviate markedly from the actual field observations. Consequently, they cannot capture the landslide’s sensitive response to rapid reservoir water level drawdown. This comprehensive comparison indicates that the actual saturated permeability coefficient of the Yanshangou landslide falls within the low range. Under such low permeability conditions, the slope stability is highly sensitive to changes in the reservoir water level.

The FOS calculation results not only validate the rationality and reliability of the slope stability evaluation method integrated with the Morgenstern–Price limit equilibrium method and Richard’s equation in this study but also provide a robust theoretical and data basis for deciphering the stability evolution mechanism of the Yanshangou landslide and formulating targeted geological disaster prevention and mitigation strategies. It is worth noting that the calculated FOS values are slightly above 1.0, indicating that the slope is globally stable but in a critical state with progressive deformation. The measured displacements represent gradual deformation along the slip zone, rather than overall failure. This is consistent with the typical behavior of reservoir landslides under reservoir water level fluctuations.

4.3. Analysis Results of Landslide-Induced Surge Hazard

Although the Yanshangou landslide has not yet occurred, based on its special geographical location in the Baihetan Reservoir area and the obvious deformation signs that have appeared, it poses a potential threat to the reservoir operations. Conducting advance analysis of landslide-induced surge hazards is of great practical significance and disaster prevention value. This study targets two typical reservoir water level scenarios (815 m and 765 m) and adopts a SPH-based numerical model to systematically simulate the entire chain process of landslide movement, surge generation, propagation, and evolution. It deeply analyzes the spatiotemporal distribution characteristics, velocity field evolution laws, and disaster intensity of surges under different reservoir water level conditions, providing a scientific basis for the early warning and targeted prevention and control of landslide-induced surge hazards.

As illustrated in Figure 12, with the progress of the landslide entering the water, the spatial distribution of surge particles presents distinct phased evolutionary characteristics. After the landslide starts (10 s, Figure 12b), the leading edge contacts the water surface, and the impact force squeezes and lifts the water particles near the contact area, forming an initial surge. The surge is concentrated near the landslide entry point, with the water surface slightly convex upward. At 20 s (Figure 12c), the landslide mass continues to accelerate into the water, and a large number of rock particles occupy the water space, causing intense compression of the water mass. The surge expands synchronously in the horizontal and vertical directions, the reservoir water level elevation at the center of the surge reaches approximately 830 m, and the influence range extends to both sides along the reservoir bank. At 30 s (Figure 12d), the landslide mass has fully entered the water, and the surge propagates further outward. The peak elevation decreases slightly to around 815 m, but the horizontal influence range continues to expand, showing a fan-shaped propagation pattern. At 40 s and 50 s (Figure 12e,f), the surge gradually attenuates due to energy dissipation, the water surface tends to be flat, and the particle distribution returns to a relatively uniform state, but local water flow disturbance still exists.

The evolution of the surge velocity field directly reflects the energy transmission and dissipation process of the landslide-induced surge, as shown in Figure 6. At 10 s (Figure 13b), the water particles near the landslide entry point are impacted by the landslide mass, generating a maximum velocity of 15 m/s, and the high-velocity area is concentrated in the local range of the contact surface. At 20 s (Figure 13c), the velocity of the surge particles reaches the maximum during the simulation period, with the peak velocity exceeding 25 m/s. The high-velocity area expands significantly, forming a dominant velocity direction consistent with the landslide sliding direction, and the velocity gradually decreases from the center of the surge to the edge. At 30 s (Figure 13d), the surge velocity begins to decrease, the peak velocity drops to about 20 m/s, and the high-velocity area continues to expand outward, indicating that the surge energy is propagating to the surrounding water mass. At 40 s and 50 s (Figure 13e,f), the surge velocity further attenuates, the peak velocity is less than 10 m/s, the velocity distribution tends to be uniform, and the surge gradually transitions to a stable state.

Compared with the 815 m reservoir water level scenario, under the 765 m reservoir water level condition (Figure 14 and Figure 15), due to the lower reservoir water level and larger water volume below the landslide entry point, the spatial evolution of surge particles shows significant differences. First, the peak surge height and velocity at the 815 m reservoir water level are higher than those at the 765 m reservoir water level. At the 815 m reservoir water level, the maximum surge height reaches about 830 m, and the peak velocity exceeds 25 m/s, while at the 765 m reservoir water level, the maximum surge height is about 795 m, and the peak velocity is about 20 m/s. This is because under the high reservoir water level condition, the water depth below the landslide entry point is smaller, and the landslide impact force is more concentrated, resulting in greater surge intensity. Second, the horizontal propagation range of the surge at the 765 m reservoir water level is larger than that at the 815 m reservoir water level. At 50 s, the influence range of the surge at the 765 m reservoir water level is about 1.2 times that at the 815 m reservoir water level. The lower reservoir water level provides a larger water volume and propagation space, making the surge spread wider. Third, the energy dissipation rate of the surge at the 815 m reservoir water level is faster than that at the 765 m reservoir water level. At the 815 m reservoir water level, the peak velocity decreases from 25 m/s to less than 10 m/s within 30 s, while at the 765 m reservoir water level, it takes about 40 s for the peak velocity to decrease from 20 m/s to less than 8 m/s. This is because the surge intensity is greater at the high reservoir water level, leading to more intense internal friction of water particles and faster energy loss.

The simulation results of landslide-induced surges under different reservoir water level conditions provide important technical support for the disaster assessment and prevention of the Yanshangou landslide. When the reservoir water level is at a high level (e.g., 815 m), the surge induced by the landslide is limited to the reservoir area and does not affect the reconstructed Liugu Town. When the reservoir water level is at a low level (e.g., 765 m), although the surge intensity is reduced, its wider propagation range may affect more areas along the reservoir bank. Therefore, the scope of the disaster should be fully considered in the disaster prevention and mitigation plan, and the protection of the entire reservoir bank area should not be ignored. In addition, the SPH-based numerical model adopted in this study can accurately simulate the entire process of landslide-induced surges, providing a reliable method for the surge hazard assessment of similar reservoir landslides.

5. Conclusions

This study comprehensively investigates the Yanshangou landslide in the Baihetan Reservoir area through field monitoring, numerical simulation, and theoretical analysis, yielding the following key conclusions:

(a)

The Yanshangou landslide exhibits obvious staged deformation characteristics that are closely coupled with reservoir water level fluctuations. The deformation rate of the landslide showed a dynamic fluctuation trend that was highly consistent with the changes in the reservoir water level. The peak deformation rate occurred during the period of rapid reservoir water level decline, and the deformation rate decreased with the rise or stabilization of the reservoir water level, which verifies that reservoir water level fluctuation is the core triggering factor for the deformation of the Yanshangou landslide.

(b)

The Yanshangou landslide has low saturated permeability, and its factor of safety presents a clear four-stage variation trend in response to reservoir water level fluctuations. Through the calibration of numerical simulation results and field monitoring data, the actual saturated permeability coefficient of the landslide is determined to be lower than 1 × 10⁻⁶ m/s. Under this low permeability condition, the factor of safety of the landslide shows a distinct four-stage variation trend with the fluctuation of the reservoir water level, which is rapid decline, slow recovery, slight drop, and steady rise. At present, the slope is in a slow deformation state and tends to stabilize with the sustained rise in the reservoir water level, but it remains highly sensitive to reservoir water level changes. Especially rapid reservoir water level drawdown can still cause a significant drop in the stability of the landslide.

(c)

Landslide-induced surges under different reservoir water level scenarios have distinct hazard characteristics, and the Smoothed Particle Hydrodynamics model can accurately simulate the entire surge evolution process. The numerical simulation results show that under the high reservoir water level scenario of 815 m, the landslide-induced surge is characterized by high height, fast velocity, and strong local destructive power, which is confined within the reservoir area and does not threaten the reconstructed Liugu Town. Under the low reservoir water level scenario of 765 m, although the intensity of the surge is reduced, its horizontal propagation range is wider, which may affect more reservoir bank areas along the Jinsha River and lead to a larger hazard coverage. The Smoothed Particle Hydrodynamics-based numerical model adopted in this study can accurately capture the key processes of landslide–water interaction, surge propagation, and energy dissipation, providing an effective technical tool for the hazard assessment of landslide-induced surges in similar reservoir areas.

(d)

Targeted disaster prevention and mitigation suggestions are put forward based on the research results for the Yanshangou landslide. On the one hand, it is necessary to strengthen the real-time and continuous monitoring of reservoir water level changes and landslide surface and deep deformation, and avoid rapid reservoir water level drawdown in the reservoir operation process to prevent a sharp drop in the stability of the landslide caused by hydrodynamic pressure changes. On the other hand, differentiated surge hazard protection measures should be adopted for different reservoir water level scenarios. For the high reservoir water level scenario, high-strength anti-surge protective structures should be set up in key areas near the landslide to resist the impact of high-velocity and high-height surges. For the low reservoir water level scenario, the hazard prevention scope should be expanded to cover the entire reservoir bank area along the river to avoid the impact of surges with a wider propagation range.

Although the current Yanshangou landslide and its induced surge do not affect the reconstructed Liugu Town, potential deeper landslides may occur under long-term reservoir water level fluctuations, so continuous monitoring and risk prevention for deeper slope sections are necessary. Future research could focus on long-term monitoring of the landslide to track its stability evolution, optimize the numerical model considering more factors (e.g., rainfall and seismic activity), and conduct quantitative risk assessment of landslide-induced surges to further improve disaster prevention and mitigation capabilities.