Prior to the impoundment of Xiluodu Reservoir, the water level of Jinsha River was about 410 m. The Yulinerzu landslide was stable and did not show any signs of deformation.

The Yulinerzu landslide started to deform after the impoundment of the Xiluodu Reservoir, which began on 4 May 2013. The reservoir level rose to an elevation of 500 m on June 6, corresponding to the front edge of the landslide. Field investigations of the Yulinerzu landslide revealed that deformation signs began to appear on June 19th, when the reservoir level reached 530 m. By June 26th, there were disturbance marks on the slope’s gravel soil and numerous secondary cracks in both the rear and front edges. The width of the rear edge crack ranged between 30 and 60 cm, with a maximum of about 1.2 m and an estimated depth of over 5 m (Figure 4A). Disturbance marks on the slope’s gravel soil were evident, and there were numerous secondary cracks in both the rear and front edges. The secondary cracks at the rear edge were generally 20–50 m long and 5–20 cm wide, with occasional steps of 5–10 cm in some areas. The secondary cracks at the front edge were relatively smaller, measuring 10–20 m in length and 0.5–3 cm in width (Figure 4B). Additionally, there were collapses along the reservoir bank.

On 31 July 2013, the reservoir level peaked at 554 m and then began to decline, dropping to 540 m by August 15th. During this drawdown period, the Yulinerzu landslide experienced intense deformation, with significant increases in crack width. The soil on both sides of the landslide collapsed and accumulated within the cracks, forming a stretching groove with a width of 5–8 m at the rear edge (Figure 4C). Secondary arc-shaped tensile cracks formed in the front parts of the slope, measuring from 3 to 5 cm up to about 10–20 cm in width. Some areas showed noticeable downward displacements, with a maximum of about 50 cm (Figure 4D). In addition, a bank collapse at the front edge of the slope increased the height by a maximum elevation of 25–30 m above the reservoir level.

On 29 September 2014, the reservoir water level reached its maximum designed elevation of 600 m and subsequently fluctuated between 600 m and 540 m. On-site investigations from 2014 to 2017 revealed that landslide deformation mainly occurred during periods of reservoir water level decline. After 2017, no further noticeable deformation of the landslide was observed. In December 2021, a combination of field investigation and unmanned aerial photography was used to determine the final distribution of cracks on the landslide, as shown in Figure 2A and Figure 5. The final height of the rear main scarp ranged between 8 and 26 m.

After the Yulinerzu landslide was found to be deforming, an automatic Global Navigation Satellite System (GNSS) consisting of nine displacement monitoring points was installed in September 2013 to monitor its surface displacement. The horizontal and vertical accuracies of the GNSS displacement measurements were 5 mm and 10 mm, respectively. Concurrently, an automatic rain gauge was installed on the landslide to monitor precipitation, while reservoir level data were provided by the Xiluodu Hydropower Project’s management.

The horizontal displacements of each GNSS monitoring point from September 2013 to December 2020 are shown in Figure 2A. Due to the submergence of the landslide’s toe, the maximum displacement was recorded at the front of the middle gentle slope platform, followed by the upper part of the landslide. In contrast, the middle and rear parts of the gentle slope platform displayed the least displacement. Generally, the horizontal displacement of each point ranged from 11,281.4 mm to 14,701.3 mm, with a displacement orientation of 217°–242°. The similarity in the displacement magnitude and orientation at each point indicated that the landslide was characterized by overall movement.

The displacement-time curves of four GNSS points (TP03-06) at the main cross section 1-1 are shown in Figure 6, accompanied by the reservoir level and daily precipitation. These data show a close correlation between the movement of the landslide and the reservoir level fluctuation. However, there is no clear correlation with precipitation. The deformation of the landslide exhibits a stepwise growth trend over time. Apart from the initial storage year, when the rise in reservoir level triggered the reactivation of the landslide, subsequent movement mainly occurred during years of reservoir drawdown.

The deformation amplitude of the landslide decreased from 2014 to 2016, and then remained relatively stable after 2018 (Figure 7). In 2014, as the reservoir level dropped from 560 m to 540 m, the annual horizontal displacements of TP03-06 ranged between 3067.9 mm and 3740.5 mm. In 2015, the reservoir level experienced its first drop from its maximum operating elevation of 600 m to its minimum operating elevation of 540 m, resulting in annual displacements ranging from 4167.9 mm to 5102.1 mm, surpassing the previous year. In 2016, the reservoir level dropped for the second time from 600 m to 540 m, with annual displacements ranging from 1283.1 mm to 1468.8 mm, significantly lower than the previous year. After 2018, annual displacements further decreased to less than 100 mm.

Since the deformation patterns across the GNSS points are similar, the relationship between landslide movement velocity and reservoir level fluctuation was analyzed using TP05 as an example (Figure 8). Temporal variations in landslide velocities from 2014 to 2017 show a rapid activation during reservoir drawdown, reaching peak velocities (25–130 mm/day). In contrast, during rest periods, velocities decreased to their minimum. Additionally, the initiation of landslide movement slightly lags behind the rapid decline of the reservoir level. This phenomenon indicates that the movement of the landslide is mainly caused by the unfavorable hydraulic gradient within the slope resulting from the drop in reservoir water level. The peak value of landslide movement speed decreased annually until 2018, after which, even during periods of rapid reservoir level decline, it was no longer significant.

As the landslide progressed towards the reservoir, its morphology changed due to differential displacements within the landslide and erosion at its front edge, ultimately affecting its overall stability. Significant displacements had already occurred from May to September 2013, predating the installation of the GNSS. Consequently, the GNSS did not capture the complete displacement of the landslide.

To comprehensively capture the morphological evolution, a multidimensional approach was employed in December 2021. A three-dimensional laser scanner was used to survey the surface morphology above the water level, while a combination of multi-beam underwater sonar and GNSS was used to survey the surface morphology below the water level. A GIS system was then used to integrate these datasets, resulting in a complete surface morphology (Figure 5). Comparing this dataset with the topographic map prior to reservoir impounding and combining it with the measured displacement vectors obtained through GNSS monitoring, the entire morphological evolution of the sliding mass was reconstructed (Figure 9).

The stability of a landslide can be influenced by toe erosion, a process characterized by bank sloughing, deformation, and retreat caused by the softening and erosion of the toe by reservoir water (Huang and Gu, 2017). However, the slope gradient of the Yulinerzu landslide front remained relatively unchanged during the reservoir operation period, as shown by the underwater topography survey results. The shear outlet of the Yulinerzu landslide, which was higher than the riverbed, led to the collapse in the section protruding from the original slope due to displacement. Consequently, the influence of toe erosion on this landslide’s stability was minimal. Overall, as the landslide progressed towards the reservoir, the main change in its geometry was the reduction in volume and a corresponding drop in the center of gravity of the upper block. In contrast, the geometry change of the lower block was very small, contributing to the overall stability of the landslide.

The combined seepage-slope stability model of the Yulinerzu landslide is presented in Figure 10, delineated along cross section 1-1' (Figure 3). The model divides the cross-section into three hydrogeological-geomechanical units, namely, slide mass, slip band and bedrock. Based on the morphological evolution of the landslide (Figure 9), three different geometries were considered: i) Model 1 corresponds to the original morphology before reservoir impoundment; ii) Model 2 represents the intermediate morphology during landslide movement corresponding to September 2013; iii) Model 3 corresponds to December 2021 and represents the landslide morphology after restoring stability.

The choice of initial and boundary conditions plays a crucial role in transient seepage modeling (Paronuzzi et al., 2013). Here, a horizontal water table level (500 m) was used as an initial condition in the seepage simulations. Variations in reservoir level during an impounding-drawdown cycle (Figure 11) were modeled through a hydraulic total head boundary condition assigned to the toe of the landslide. According to the actual operation of the reservoir, the rate of rise and fall of the reservoir water level was taken as 1.5 m/d. To reduce calculation time, the duration of the reservoir water lever at its highest and lowest points was shortened.

Transient seepage simulation requires the characterization of materials under unsaturated conditions (Pinyol et al., 2012). The Fredlund and Xing methods (Fredlund et al., 1994a, b) were used to develop the volumetric water content functions and hydraulic conductivity functions. The physical and hydrogeological properties adopted for the three hydrogeological units are indicated in Table 1 and the corresponding hydraulic functions are shown in Figure 12. The shear strength parameters were back-calculated based on the stability and observed failure process, with initial trial values determined from laboratory tests.

The groundwater table variation caused by the reservoir level fluctuation is depicted in Figure 13. As the reservoir level changes, the groundwater table in the permeable sliding body is almost synchronized with the reservoir level. However, due to the low permeability of the slip band, the variation of groundwater level in the bedrock below the sliding zone lags behind that of the sliding body, creating a water head difference above and below the slip band (Figure 11). During the reservoir impounding period, the water head above the slip band is higher than below it, resulting in seepage pressure acting inward on the slope. This imparts some slope stability. Conversely, during the drawdown period, the water head above the slip band remains higher than that below it, leading to excess pore water pressure below the slip band. This condition is not conducive to slope stability.

Given the negligible impact of rainfall on the deformation of the Yulinerzu landslide, this paper only considered the influence of reservoir level changes on the stability of the landslide. The pore water pressure obtained from the seepage calculations was introduced into the limit equilibrium slope stability analyses using the Morgenstern-Price method and the SLOPE/W code (Geo-Slope International Ltd, 2007). The sliding surface, aligned with the slip band based on the geological interpretation, is governed by the widely used Mohr–Coulomb failure criterion, with strength parameters shown in Table 1.

The variations in stability conditions of the three different geometric models (Figure 12) throughout an entire impounding-drawdown cycle were calculated by combining seepage and slope stability analyses (Figure 14). The variation patterns of FOS with the reservoir level are similar for the three geometric models. The FOS decreases during the impoundment period and increases during the drawdown period. From the geometric shape, under the same reservoir level conditions, the FOS of Model 1 is the smallest and that of Model 3 is the largest.

Considering the dynamic evolution of geometric morphology, the slope stability trend reflected in the calculation is consistent with the actual situation of the Yulinerzu landslide during the impounding-drawdown cycle. Model 1 reflects the initial impoundment of the reservoir to an elevation of 554 m (until September 2014). When the water level rose to 530 m, the FOS of the landslide decreased to 0.997, and deformation of the landslide began to occur. Model 2 corresponds to the period from September 2014 to 2018. During this stage, except when the reservoir level decreased from 600 m to 570 m, the FOS of the landslide was less than 1. At other times, the FOS of the landslide was greater than 1. Therefore, the deformation of the landslide mainly occurred during the period of water level decline. Model 3 reflects the situation after 2018, when the FOS of the landslide was greater than 1, indicating that the landslide returned to a stable state.

The hydraulic effects of reservoir level changes on landslide stability can be divided into two aspects (Tang et al., 2019). The first is the floating effect. As the water level in the reservoir rises, the volume of the submerged slide mass increases, leading to an increase in water pressure on the sliding surface and a decrease in effective stress. Conversely, when the water level drops, the opposite occurs. The second effect is the seepage pressure effect, which generates a seepage pressure inward on the slope when the reservoir level rises and outward when it drops. These two effects have opposing impacts on landslide stability, and their relative magnitude determines the final impact on landslide stability.

The floating effect and the seepage pressure effects are determined by the rate of reservoir level change, hydrogeological structure and hydraulic parameters of the landslide. Tang et al. (2019) analyzed and discussed the activity and displacements of reservoir landslides with special regard to different sliding surface and permeability and presented four dominant combinations of the inducing factors of re-active landslide. In their research, the hydrogeological structure of landslide was simplified as slide mass and bedrock. In the study of Canelles landslide, Pinyol et al. (2012) found that the hydraulic behavior of the clay layers where the sliding surface is located appears to be independent of the underlying sandstone. Due to the difficulty in dissipating or increasing water pressure within the clay, the failure of Canelles landslide is a result of rapid drawdown and the low water level reached in the reservoir. The Yulinerzu landslide is a colluvial landslide with a complex hydrogeological structure, characterized by a high permeability coefficient in the slide mass and a low permeability coefficient in the slip band, while the permeability coefficient of the bedrock below the slip band is relatively high. When the reservoir level changes, the groundwater level in the slide mass synchronizes with the reservoir level, and the seepage pressure is mainly manifested as the water head difference above and below the slip band. When the reservoir level rises, the adverse impact of the floating effect is greater than the beneficial impact of the seepage pressure effect, reducing the FOS of the landslide. Therefore, the floating effect is the main cause of deformation of the Yulinerzhu landslide during the initial impoundment. As the reservoir level continues to rise, the landslide regains its equilibrium due to its own morphological adjustment.

During the period of reservoir level decline, the adverse impact of the seepage pressure on landslide stability becomes greater than the beneficial impact of the floating effect, resulting in the lowest safety factor for the landslide and further deformation. This explains why the deformation of the Yulinerzu landslide lags behind the reservoir level decline. However, as the water level further decreases, the beneficial impact of the floating effect exceeds the adverse impact of seepage pressure, and thus the FOS of the landslide begins to increase, leading to a resumption of stability.

The geometry of a landslide, especially the shape of its sliding surface, is an important factor affecting the deformation patterns, and failure behaviors, and their overall evolution. Sang et al. (2020) indicated that the strain distributions depend on the position of the main slip surface within the sliding zone, as well as the shear deformation of the sliding zone. Based on the difference in the dip angle of the sliding surface, the slide body of the Yulinerzu landslide was conceptually simplified into two blocks: an upper block with a sliding surface dip angle of 44°, and a lower block with a sliding surface dip angle of 12° (Figure 12). The upper block had a larger dip angle of the sliding surface, giving it a greater potential energy. When the resistance force of the lower block decreased under the action of the reservoir’s water, the landslide manifested a tendency for revival deformation. Under the push of the upper block’s high potential energy, the slide body moved forward as a whole, causing a proportional reduction in the upper block’s potential energy. With the repeated rise and fall of the reservoir water level, the slide body was repeatedly activated and braked. This cycle weakened the potential energy of the upper block, gradually decreasing the overall deformation rate of the landslide decreased, until it stabilized.

Brunsden (2001) indicated that a landscape change takes place as a process–response function. Tang et al. (2015) investigated the evolutional characteristics of the Huangtupo landslide in the Three Gorges Reservoir region of China based on in situ tunneling and monitoring. Chen et al. (2023) analyzed the spatial distribution and failure mechanism of water-induced landslides in the reservoir areas of Southwest China. They found that as displacement accumulates, the movement of the landslide mass is a self-stabilizing process, however, they did not explain this mechanism. Based on the measured morphology evolution process of the Yulinerzu landslide, this paper used a limit equilibrium analysis method to analyze the safety factor of the landslide at different stages and obtained calculation results consistent with the actual situation.

Therefore, the morphological evolution caused by the accumulation of displacement is an important reason for the self-organizing characteristics of landslides. It is important to note that as the morphology of a landslide changes, its hydrogeological properties also change, which can affect its stability. This issue necessitates further research.