Identification and Mitigation of Reservoir Landslides: Cases Studied in the Three Gorges Reservoir Area of China

Abstract

Reservoir landslides are one of the most common types of geohazards in reservoir areas. It is of great theoretical significance and engineering application value to carry out the identification and mitigation research of them. In the Three Gorges Reservoir area (TGRA), a series of studies were conducted on the reservoir landslides in aspects of their characteristics, evolution mechanisms, prevention and control, and forecasting by means of field investigation, monitoring and in-situ experiment, etc. Massive corpora of these research work are summarized and presented here in five aspects, with the goal of gaining insights into the effective prevention and mitigation of them: (1) the characteristics of the reservoir landslides located in the TGRA were summarized; (2) the stability evaluation methods of reservoir landslide include qualitative, limit equilibrium methods, numerical simulation methods and probabilistic methods were introduced; (3) the in-situ experiment on slipping zone soil of reservoir landslides carried out on the Huangtupo landslide was described; (4) the research progresses in reservoir landslide prevention and control was introduced; (5) the prospects of reservoir landslides forecasting based on physical-mechanical mechanism was elaborated. Based on the overview, suggests topics for future research are given to address the remaining challenges of reservoir landslides.

1 Introduction

Landslides caused by the impoundment and drawdown of reservoir are called reservoir landslides. They’re widespread throughout the globe, posing a threat to the safety of residential areas, roads, and water transportation in the reservoir area. Once it occurs, on the one hand, the impact of the sliding mass can produce huge surges, which causes flood when crossing the dam and rushing to the downstream; On the other hand, the sliding mass can block the river and form a natural dam, causing upstream backwater and potential threats of debris flow or flooding. The 1963 Vajont landslide is a typical reservoir landslide in Italy. The sliding mass glides down as a whole at a maximum speed of 30 m/s, causing a surge of 250 m high that demolished the Longarone town and caused over 2000 deaths (Paronuzzi and Bolla 2012). Reservoir landslides is a serious natural disaster that must attract attention.

The influencing factors of reservoir landslides are various, including geological environment, rainfall, reservoir water level fluctuations, and engineering excavation, etc. (Huang and Gu 2017; Wen et al. 2017; Froude and Petley 2018). Water is a key factor in all of these. Through the affection of hydrostatic pressure, buoyancy force, hydrodynamic pressure, the sliding zone is developed and the physical-mechanical properties of the sliding mass are changed, which eventually lead to the landslide. Therefore, figure out how to reduce the impact of water has always been the top research priority in the mitigate of reservoir landslides. Although hundreds of billions of RMB have been invested in the study and treatment of these reservoir landslides, many of them remain in a state of continuous deformation, with cumulative displacements of several meters now documented at the Huangtupo Landslide (Tang et al. 2015a; Zou et al. 2020; Dumperth et al. 2016), Outang (Yin et al. 2016), and Baishuihe (Li et al. 2010; Du et al. 2013).

The identification and mitigation study of reservoir landslides has attracted wide attention globally, as a great deal of scientific research has been undertaken on them over the past decades. Statistical data show that, among the 969 papers on the theme of reservoir landslides from 1999 to 2018 included in the Web of Science (WOS), only a few were published annually before the completion of the Three Gorges Dam (TGD) in 2003. Then, the number of papers related to reservoir landslides and the number of citations increased rapidly at an annual rate of 20–40% (Li et al. 2020). The construction of the TGD in China has greatly enriched the study of reservoir landslides.

The VOS viewer’s examination results of the abstracts and titles of 969 papers aforementioned are displayed in Fig. 1. Each ball’s size denotes the frequency with which that term appears, the line thickness between balls represents the degree of link between two high-frequency terms, and the balls’ colors represent the many topics to which those terms pertain. Results show that research on reservoir landslides over the past 20 years has primarily focused on the three topics as shown by the distribution of high frequency terms related to reservoir landslides in Fig. 1. These topics are “deformation, stability, seepage” (red ball), “prediction, landslide hazard, network” (blue ball), and “sediment, erosion, deposit” (green ball). Density plots of these frequently-used terms reveal that “China,” “Gorges Reservoir,” “deformation,” and “stability” are the most-used and most-related terms, suggesting that the stability of the reservoir landslide in the Three Gorges Reservoir area (hereafter referred to as TGRA) is the primary focus of reservoir landslide study.

Fig. 1

High frequency terms in title and abstract from 1999 to 2018 (after Li et al. 2020)

To make a comprehensive entirety introduction of the reservoir landslides, the study of reservoir landslides in the TGRA was summarized here from the following five aspects: (1) characterizing TGRA reservoir landslides; (2) evaluating reservoir landslides’ stability; (3) conducting of reservoir landslides in-situ experiment; (4) designing of reservoir landslides prevention and control measures; and (5) elaborating the potential for reservoir landslide forecasting based on physical-mechanical mechanisms. Suggestions for future research toward the goal of disaster reduction and discusses difficulties engineering geologists face in the prevention and mitigation of TGRA reservoir landslides were also given (Tang et al. 2019).

2 Characteristic of Reservoir Landslides in the TGRA

2.1 Spatiotemporal Distribution

Along the Yangtze River, the distribution density of landslides shows clearly distinct regional differences. The landslides tend to concentrate along the Yangtze River and some of its tributaries but are most common to the east of Wanzhou in the TGRA, (Fig. 2a). In terms of quantity, landslides and rock avalanches are especially common in Zigui and Badong counties (Fig. 3). Considering the tributary rivers, landslides are most frequent in the catchments of Xiangxi, Guizhou, Qinggan, Caotang, Meixi, and Wujiang Rivers, accounting for 44.3% of the total number of landslides and collapses present along the Yangtze River tributaries.

Fig. 2

The corresponding relationship between the river terrace elevation and the occurrence of the landslide (after Tang 2015; Tang et al. 2019; Tian et al. 1996)

Fig. 3

Number of landslides and collapses along the impounded reach of the Yangtze River starting from the westernmost area of Jiangjin (JJ) to the easternmost area of Yichang (YC). (See Fig. 2 for location) (after Tang et al. 2019)

The evolution process determines the complexity and diversity of reservoir landslide, and also determines the boundary conditions and spatiotemporal development law of landslide. For instance, it is found that the occurrence time and front elevation of landslide have close relationships with the neotectonic movements in different periods of the Yangtze River and the elevation of river terraces (Fig. 2). The formation of slip zone is controlled by multi-stage crack expansion and connectivity, revealing the multi-scale spatial distribution law of reservoir landslide.

The landslides in the TGRA generally have significant regularity in the elevation distribution. The elevations of the front edges of reservoir landslides that occur along the Yangtze River are shown in Fig. 2b. Among them, 93% of their front edges’ elevations are concentrated between the elevation segment of 100–175 m, which was the water level fluctuation zone after the construction of the TGD. This indicates that the reservoir landslides are closely related to the effects of the fluvial erosion of the slope toe. The river water erodes the side part of riverbed or the slope toe, gradually forming better free space conditions, which makes the upper sliding mass lose sufficient support and leads to shear failure. Overall, the reservoir landslides are concentrated in the first and second terraces as well as reservoir water level fluctuation zones, while only a small amount of them are developed in the third terrace. The number of reservoir landslides decreases with increasing elevation. This is due to the relatively late formation of reservoir landslides in the first and second terraces, while those that formed earlier in the third terrace were gradually denuded or even disappeared during their evolution process under the influence of reservoir (Li et al. 2018).

2.2 Slide-Prone Strata

Lithology is a primary determinant of reservoir landslides distribution in the TGRA (Li et al. 2018). The sandstone, mudstone, and interbedded sandstone/mudstone layers of the Triassic Badong Formation and the Jurassic strata are known as the most slide-prone strata, producing numerous landslides. The reservoir landslides identified in these strata account for 91.1% of the total landslide volume in the TGRA.

The Badong formation strata are distributed widely in the TGRA, in which large-size landslides and deep-reaching loosely consolidated formations are likely to occur. The TGRA’s Badong formation strata can be divided into five sections (from T²b¹ to T²b⁵), with lithology combinations characterized by sandwich of soft and hard strata. The lithology and lithological combinations exhibit a unique spatial law of variation. The strata of the Badong formation are characterized by rich variation, an outcome of long-term slope deformation as well as all-important evidence for understanding the triggers and mechanism of landslide formation.

Figure 4 shows the histogram of strata measured. Generally, and it is characterized by thick stratum in the east and thin stratum in the west. The horizontal stratum thickness of the Badong formation is unstable and subject to great variation. The spatial differences of stratum thickness of the Badong formation are attributed to three factors: different thickness of deposition, uneven denudation, and impact of subsequent structural deformation.

Fig. 4

Histogram for comparison of the strata measured in Fengjie–Wushan Mountain–Badong–Zigui of the Badong formation

No large landslides are associated with Precambrian magmatic and metamorphic rocks, and only small slumps occurred in the weathered materials near the dam. Additionally, slope instabilities are relatively infrequent in the carbonate rock area that hosts the famous Three Gorges. Less than 8% of the total volume of identified landslides is associated with these rocks.

2.3 Influence of Geologic Structure

In the TGRA, landslides are most likely to occur where strata drop at moderate angles toward rivers, specifically on the flanks of anticlines and cores of synclines. These settings are suitable for the formation of large consequent bedding landslides. Two notable examples are the Qianjiangping and Huangtupo landslides, which originated in the Zigui and Guandukou synclines, respectively (Fig. 2a). Reservoir landslides in the TGRA can be attributed to six main evolutionary models as is shown in Fig. 5, in which, the geologic structural and evolution characteristics are stand out a mile.

Fig. 5

Evolution modes of reservoir landslides. (a) Progressive slip along gentle-dip layer; (b) Bucking failure along steep-dip layer; (c) Creep slip along deep layer; (d) Plastic flow slip of weak interlayer; (e) Breakthrough abrupt slip in inclined cross-cutting layer; (f) Bending slip in steep anti-dip layer

2.4 Dynamic Characteristics of Reservoir Landslide Evolution

• (1) Deformation process of landslides

The deformation and failure processes of reservoir landslides were studied via site surveys, monitoring, physical model tests and numerical simulations. Many TGRA landslides were affected by periodic variations in water level and rainfall. Some landslides deform intermittently and show step-like cumulative displacement curves, such as the Baishuihe landslide (Li et al. 2010), Shuping landslide (Wu et al. 2019) and Outang landslide (Yin et al. 2016). Other TGRA landslides deformed continuously and had linear-like curves of cumulative displacement, such as the Majiagou landslide (Ma et al. 2017).

Researchers have utilized a variety of techniques in studying and reconstructing the evolution process and deformation mechanism of reservoir landslides, such as physical model experiments and numerical simulations. Others adopted finite element analysis to examine the coupling effect of precipitation and reservoir water levels variation. (Jiang et al. 2011; Zhao et al. 2017). The Anlesi landslide mechanism was studied using a rheology model (Jian et al. 2009). The effects of rainfall and water level fluctuations on the behavior of the Quchi landslides were investigated using a universal discrete code (Huang and Gu 2017).

• (2) Mechanical characteristics of reservoir landslides

The stability of the landslides affected by the reservoir water mainly via hydrostatic pressure, buoyancy force, and hydrodynamic pressure, etc. Based on the different affection way of water in the reservoir landslides’ deformation, the mechanical characteristics of reservoir landslides could be classed as seepage-induced or buoyancy-induced. In seepage-induced landslides, deformation occurs mostly when reservoir water level decreases (Fig. 6a), but in buoyancy-induced landslides, deformation occurs primarily when reservoir water level is high (Fig. 6b). Landslides triggered by seepage are more likely to occur in sliding material with low permeability. When the reservoir water level drops abruptly, it is difficult for pore water pressure to dissipate, resulting in outward seepage force that tends to destabilize the landslide. Buoyancy-induced landslides are more likely to arise in sliding masses with a high permeability. The reservoir water penetrating the sliding material as the water level rises softens the sliding zone and lightens the sliding mass, reducing the landslide’s stability. These two forms of mechanical characteristics are exemplified by the Baishuihe and Muyubao landslides as is shown in Fig. 6c, d. Two typical reservoir landslides of these two types, Shuping landslide and Qianjiangping landslide, are detailed described in Sects. 2.6 and 2.7.

Fig. 6

Landslide deformation can accelerate by either rising or falling reservoir water level, depending on whether seepage forces (parts a, c) or buoyancy forces (parts b, d) are predominant. The differences between the seepage-driven Baishuihe landslide (c) and the buoyancy-driven Muyubao landslide (d) are observed through the corresponding periods of faster movement (red boxes) (after Tang et al. 2019)

2.5 Influence of Rainfall and Fluctuating Water Level

Reservoir landslide’s occurrence is usually triggered by the interaction of external factors and specific regional geological conditions. Rainfall and reservoir water level fluctuations are considered to be the two major external influencing factors of reservoir landslides (Tang et al. 2019).

• (1) Influence of rainfall

Seasonal rainfall is a significant factor that affects reservoir landslides (Wang et al. 2021a). For example, the rainfall in the TGRA is heavy during the annual low water level period. Rainwater infiltrates the sliding mass, increasing its weight and raising the groundwater level, which reduces the effective stress of the slide zone and causes the landslide to be unstable. It has also been noted that rainfall at the rear of large reservoir landslides significantly raises its groundwater level, increases the head difference between the leading and trailing edges of sliding mass, significantly increases the hydraulic gradient, and is extremely detrimental to the stability of landslides (Zhang et al. 2021a).

The study of how the rainfall affects the reservoir landslides in TGRA varies. Water infiltration mechanisms and rainfall’s impact on slope stability were studied by multiple researchers with field monitoring. Observation wells were constructed in typical colluvial landslides (Tang et al. 2015b) and decomposed granite slope (Zhang et al. 2000). The rainfall infiltration process was studied using the monitoring data on suction and temporary pore pressure variations during the rainfall. Additionally, the influence of rainfall characteristics (amount, intensity, distribution pattern) on slope stability was studied (Yang et al. 2017; Liu et al. 2018).

• (2) Influence of reservoir water level fluctuation

Reservoir impoundment has clearly increased the frequency of reservoir landslides in the TGRA. Most significant was the reactivation of large landslides, whose newly submerged toes became subject to fluctuating hydrologic conditions (Song et al. 2018).

Seepage, in the form of hydrodynamic pressure, is considered to be the primary reactivation mechanism of most (70%) landslides, whereas 15% of the landslides were affected by buoyancy in the TGRA (Yan et al. 2010). When the reservoir water level drops rapidly, the water level drawdown inside the landslide lags behind due to the poor permeability, which causes hydrodynamic pressure and seepage-induced landslides to form under these circumstances. When the reservoir water level rises and remains at a certain height, the weight of the foot-of-slope rocks diminishes under the effect of the buoyancy force, the passive forces decrease, and the stability is disturbed (Lindenmaier and Zehe 2005). Buoyancy-induced reservoir landslides generally occur under this condition.

The first storage of the reservoir is considered to be the most hazardous. During the initial storage period of the TGRA, the effect of reservoir water was strong and long-lasting, e.g., the number of newly formed landslides has been at a high level for 3 years after 135 m of storage. This effect has since diminished year by year; for example, the number of newly formed landslides has dropped rapidly to a very small value after 1 year of 175 m storage.

The periodic TGRA water level fluctuation can alter the seepage field, the stress field and the material properties, resulting in a change in slope stability. Such findings were reported by Hu et al. (2012), Zhang et al. (2014a) and Zhao et al. (2017). The effect of rising and falling water levels was also studied through model tests or numerical simulations considering different rates of water level variation (e.g., Sun et al. 2017a, b). In addition, Liao et al. (2005) demonstrated that slope stability was closely related to the variation in the permeability coefficient of rocks.

• (3) Combined effect of rainfall and water level fluctuation

In many instances, the development of reservoir landslides depends on the combined effect of rainfall and water level fluctuation. Numerical simulations were used to study the characteristics of the seepage field based on saturated-unsaturated theory (Miao et al. 2018). Many TGRA landslides have been continuously monitored. Based on the monitoring data, the influence of rainfall and water level fluctuation on the landslide deformation was analyzed (Li et al. 2010; Tang et al. 2015a; Wang et al. 2017). In addition, data mining methods, such as gray relational grade analysis, two-step cluster analysis and Apriori algorithm analysis, were applied to examine and quantify the influence of rainfall and water level variation on landslide deformation (Ma et al. 2017; Tan et al. 2018a).

For a better understanding of the combined effect of rainfall and water level fluctuation on landslide’s deformation, a correlation analysis between displacement velocity at G1 of Shiliushubao landslide and its rainfall, rate of reservoir level change, and reservoir level are shown in Fig. 7. The size of the bubbles represents the deformation velocity. The larger bubbles tend to plot where the rainfall is higher. Meanwhile, the large bubbles are mainly concentrated where the reservoir level is between 140 and 150 m and are located where the reservoir level fluctuates slowly (between −4.4 and 9.0 m/month). This indicates that reservoir level fluctuations mainly trigger accelerated landslide movements when the reservoir level is low. The maximum size bubble appears where the rainfall is about 325 mm/month, and the water level rises between 4.5 and 9.0 m/month. The combined effect of heavy rainfall and rising reservoir level on landslide deformation is more significant than low rainfall combined with reservoir level drawdown.

Fig. 7

Correlation between displacement velocity, reservoir level, rainfall, and fluctuation of reservoir level at Shiliushubao landslide (after Zhang et al. 2021b)

2.6 Seepage-Induced Reservoir Landslide

The Shuping landslide is a typical seepage-induced reservoir landslide. It is located in Shazhenxi Town, Zigui County, Hubei Province, 47 km upstream from the TGD, and on the southern bank of the Yangtze River (Fig. 8). After the first impoundment of the reservoir in 2003, significant deformation was observed, endangering 580 residents and Yangtze River navigation. Therefore, multiple monitoring techniques were utilized to assess the deformation characteristics and stability of the landslide, as well as to observe the interactions between the various landslide segments (Fig. 9).

Fig. 8

Location and photograph of the Shuping landslide (after Zhang et al. 2020)

Fig. 9

Monitoring arrangement of the Shuping landslide (after Zou et al. 2021)

Professional monitoring of surface displacement commenced in June 2003, with seven GPS monitoring points (SP2, ZG85, ZG86, ZG87, ZG88, ZG89, and ZG90) (Fig. 10). The system was upgraded to automatic, real-time monitoring in June 2012, and all GPS monitoring points were surveyed every half month. The daily rainfall and daily reservoir level are recorded by Meteorological Station near the Shuping landslide and China Three Gorges Corporation, respectively.

• (1) Overall deformation feature

The results of monitoring indicate that the evolution of the Shuping landslide consists of four stages that related to human activity, which are 139-m above-sea-level (ASL) trial reservoir impoundment (from April 2003 to September 2006), the 156 m ASL trial reservoir impoundment (from September 2006 to September 2008), the 175 m ASL trial reservoir impoundment and deformation control engineering project of the Shuping landslide conducted between September 2014 and June 2015 (Fig. 10).

Fig. 10

Monitoring data for Shuping landslide from 2003 to 2016 (after Zou et al. 2021)

The GPS monitoring system and field investigations revealed the deformation features (Figs. 10 and 11). The main sliding mass can be divided into a main deformation area and a secondary deformation area (Fig. 9). As measured at sites ZG85, ZG86, ZG88, SP2 and SP6, the main deformation area underlies most of the area with the cumulative displacement up to 4–5 m. Throughout the 13-year monitoring period, point SP2 experienced the greatest cumulative displacement (5.168 m), followed by ZG86 and ZG88, which recorded 5.039 m and 4.919 m, respectively. Deformations at the monitoring sites were essentially synchronous, as evidenced by the similar shape of their cumulative displacement curves, which typically exhibit steady rises in the first impoundment stage, step-like trends in the second and third impoundment stages, and flat trends after the engineering treatment. In the secondary deformation area, deformations were smaller and more stable, as indicated by the gentle cumulative displacement curves at ZG89, ZG90, and ZG87, which recorded cumulative displacements of 0.5–2 m from 2003 to 2016.

• (2) Deformation feature in different stages

After the reservoir level rose to 135 m ASL for the first time in June 2003, the main deformation area deformed at an average rate of 15.6 mm/month until September 2006, with each site recording relatively steady displacement curves whose small or nonexistent steps correspond to the annual variations in reservoir level. During Stage 1, there was no obvious deformation at ZG89 and ZG90 in the secondary deformation area.

Fig. 11

Time series of reservoir level, rainfall and landslide displacement from 2003 to 2016. (a) Reservoir water levels and variation rates (positive for level rise, negative for level drop); (b) Displacement velocity of the GPS points in the main deformation area and monthly rainfall; (c) Displacement velocity of the GPS points in secondary deformation area and monthly rainfall (after Zou et al. 2021)

When the reservoir level first rose to 156 m ASL during the first 2 months of Stage 2 (September and October 2006), displacement velocities of the main deformation area decreased to 13.4 and 9.7 mm/month, respectively, indicating an improvement in sliding mass stability. During the subsequent 2 months, the velocity increased to 11.5 and 14.3 mm/month while the reservoir level remained unchanged at 156 m ASL. During the subsequent drawdown period, the displacement velocity increased to a maximum of approximately 100 mm/month when the reservoir level dropped to 145 m ASL in 2007 (Fig. 12), resulting in an average “jump” of 458 mm in the cumulative displacement curve, which then flattened while the reservoir level remained at 145 m ASL (Fig. 10).

Fig. 12

Monitoring data of GPS point SP2 on the middle part of the Shuping landlide, from December 2013 to September 2014 (after Zou et al. 2021)

During the beginning of Stage 3, when the reservoir first rose to nearly 175 m in October 2008, the displacement velocity of the main deformation area decreased to 12.7 mm/month compared to 65, 74, and 32 mm/month in the previous 3 months. Shortly after the reservoir rose to its highest level, the level underwent a gradual decline, and the displacement velocity increased steadily. The maximum displacement velocity reached 378.6 mm/month at ZG88 in May 2009 when the water level declined rapidly, a rate almost four times higher than when the reservoir dropped from 156 to 145 m ASL in 2007. After the water level was steady at 145 m ASL, the displacement velocity decreased to a relatively low value (Fig. 11b).

During the next 6 years of Stage 3, the reservoir level underwent a series of similar annual fluctuations, and the sliding mass responded with a series of “jumps” in deformation. During these cycles, the displacement velocity decreased as the reservoir level rose, remained low speed when the reservoir level remained high, increased as drawdown began, and reached a maximum of 165 mm/month during rapid drawdown. As the reservoir was maintained at a low level of 145 m ASL, the corresponding cumulative displacement curves flattened considerably. These results clearly demonstrate that displacement velocity is high during reservoir drawdown and low during reservoir rise.

After the engineering treatment was completed in June 2015, the cumulative displacement curves became extremely flat and the “jumps” disappeared (Fig. 11). In the main deformation area, the deformation was reduced to a low level of 4.1 mm/month, demonstrating effective treatment.

• (3) Effect of water-level fluctuation and rainfall on the deformation of Shuping landslide

The largest “jump” in the cumulative displacement curves occurred in May to June of 2012 with an average of 479 mm and the second largest “jump” of 458 mm in May to June of 2009. These periods corresponded to the two highest monthly drawdown rates of 9.67 and 9.38 m, respectively (Fig. 11a). During these 2 years, the highest monthly precipitation amounts were 180 mm/month in 2009 and 190 mm/month in 2012 (Fig. 11). These data demonstrated conclusively that the deformation of the Shuping landslide is driven primarily by reservoir level fluctuations and not by precipitation. Low displacement velocities and flat cumulative displacement curves during the peak of the rainy season in July and August, when the reservoir is at its lowest level, also confirm this relationship.

The influence of reservoir level and rainfall on landslide deformation is illustrated in Fig. 12. The reservoir level fell at an average rate of 0.041 m/day in December 2013, and the corresponding velocity of displacement was 0.22 mm/day. In the subsequent 3 months, the rate of drawdown of the reservoir level increased to 0.147 m/day and the velocity of displacement increased to 0.54 mm/day. In March 2014, the displacement velocity decreased as the water level rose, despite the fact that heavy rainfall (up to 27.5 mm/day) was recorded during this time. During the subsequent rapid drawdown period (0.419 m/day) from May to June, the velocity of displacement increased to approximately 5 mm/day. Subsequently, the velocity of displacement decreased to less than 1.2 mm/day as the water level remained low despite abundant precipitation. These facts demonstrate that the Shuping landslide’s displacement velocity is positively correlated with the reservoir’s drop rate, while rainfall has little effect.

In contrast to the flat displacement curves and low displacement velocity observed in other years when the reservoir level was stable at its lowest annual level in July and August, displacement velocities were large in 2008 and 2010 (65.0 and 73.8 mm/month in July and August 2008; 58.4 mm/month in July 2010, approximately half of the average highest monthly displacement velocity, 165 mm/month, during rapid draw down period). During those periods, up to 300 millimeters (mm) of rainfall per month was recorded, which is heavy. However, August 2011 had the next heaviest rainfall of 250 mm/month, yet the cumulative displacement curve remained flat, and the displacement velocity was low (22.2 mm/month). These data illustrate that heavy rainfall can decrease landslide stability and accelerate deformation however it was a secondary factor. The difference in the displacement velocity between the months with the highest (2008, 2010) and the second highest (2011) levels of rainfall suggests that a threshold exists, with rainfall exceeding this value having a significant effect but with less having little significance. This threshold appeared to be approximately 250–300 mm/month.

2.7 Buoyancy-Induced Reservoir Landslide

The Qianjiangping landslide is a typical buoyancy-induced reservoir landslide. It is located in Shazhenxi Town, Zigui County, Hubei Province, on the western side of Qinggan River (a tributary of the Yangtze), 50 km upstream from the TGD (Tang et al. 2017). The Qianjiangping landslide formed along weak bedding planes in Jurassic strata that generally dip approximately 30 degrees to the southeast, downslope and toward the river. This mass has a length of 1150 m, a width of 600 m, and an average thickness of 30 m (Fig. 13). The total area is greater than 0.69 km² and the volume is 21 × 106 m³. Maximum elevation of the main scarp is 450 m, but it extends to river level (Fig. 14, red dotted line). On July 14, 2003, the sliding mass moved approximately 250 m toward S45°E, all within a period of 1~5 min. The slide mass is thinner in the main body and thicker at the toe. The dip angle of the lower rupture surface is mostly coincident with bedding in the upper part, however it becomes nearly horizontal at the toe of the landslide. Case histories show that slope failure in this region is common where the Jurassic strata have steep, downslope dips toward major rivers.

Fig. 13

Topography of the Qianjiangping landslide after 2003 (after Tang et al. 2017)

Fig. 14

Front view of the landslide, looking NW (after Tang et al. 2017)

The landslide is bordered on the east, south, and west by a prominent incised meander in the Qinggan River (Fig. 13). The water level of the Qinggan River rose from 95 m to 135 m in June 2003, immediately after the impoundment of the TGRA, which is just a few weeks prior to the landslide event.

• (1) Geology and structure of the landslide

The Qianjiangping landslide is a bedding-controlled landslide. The local stratigraphic section is, from top to bottom, alluvium, clay and gravel of Quaternary age, feldspathic quartz sandstone of Upper Jurassic–Cretaceous age, and fine sandstone with carbonaceous shale, siltstone with mudstone, and silty mudstone of the Middle to Lower Jurassic Niejiashan formation (J_1–2n). The sandstone is interbedded with weak silty mudstone and shale layers. The greater region is structurally complex and tectonically active. The rapid incision of the Yangtze River, thought to have occurred in response to Quaternary tectonic uplift (Li et al. 2001), has produced unstable slopes along its banks in numerous places. Field investigations show that fractures and discontinuities are abundant in the slightly weathered bedrock. The dip direction of the strata is reversed near the Qinggan River.

• (2) Hydrology and Hydrogeology

The Qinggan River has numerous tributaries and ravines. The gullies are usually dry, but carry significant intermittent flows during the rainy season. The north bank of the Qinggan River is a consequent slope with a dip angle between 20° and 30°, while the south bank slope is an obsequent topographic surface.

Near the landslide, data of groundwater are hard to obtained. The data obtained from the boreholes on the landslide indicate that the phreatic zone within the slope material is in a very deep place. Along planes of discontinuities and contacts between the dipping sandstone and mudstone units, artesian contact springs were observed. Groundwater recharge and levels are primarily determined by the amount of rainfall, the infiltration rate and the level of reservoir water. Additionally, boring revealed that a highly fractured, 1–2 m thick zone of bedrock immediately underlies the main slip zone, while the deeper bedrock was only slightly fractured. This indicates that the bedrock immediately below the slip zone has good permeability.

• (3) Inducing factors of the landslide

The monitoring report shows that, the Qianjiangping landslide occurred immediately after a nearly continuous period of rainfall from June 21 to July 11, 2003. In addition, the water level of the Qinggan River rose rapidly from 95 m to 135 m ASL shortly after the Three Gorges reservoir was impounded on June 1, 2003. Studies indicate that this landslide was caused by the combined effect of this reservoir’s water level rise and heavy rainfall (Jian et al. 2014). Groundwater data for the area surrounding the Qianjiangping landslide and its environs are limited, but it is evident that the rapid increase in the reservoir level affected the groundwater level near this landslide’s toe, which further contributed to the mass’ destabilization. The erosion at the slope toe caused by the river also decreased the slope’s stability.

Overall, the Qianjiangping landslide was caused by both internal and external factors. Internal factors include the loose accumulation of the slide mass, significant geomorphic features, and weak bedding surfaces. External factors include the precipitous rise in the reservoir’s water level in June 2003, which coincided with the season of heavy rainfall.

• (4) Mechanism analysis of Qiangjiangping landslide

Buoyancy played a vital role in the evolution of Qiangjiangping landslide. With the effect of rising reservoir water level and the rainfall, water gradually seeped from the slope surface into the interior of the sliding mass, and the water capacity within the slope varied in different reservoir filling stages. Cao and Luo (2007) and Li et al. (2013) found that the shear strength and deformation properties of slip zone soils of the Qianjiangping landslide were highly influenced by the variations in soil moisture through laboratory investigations. The reservoir impoundment affected the toe of the Qianjiangping landslide mainly, and the effect of shear strength reduction at the foot of the sliding mass was the most important factor in the landslide initiation. It was determined that the shear strength weakening of the slip zone in the lower layers of this landslide had the greatest effect on its stability (Wen et al. 2008). Specifically, the lower layers were gradually saturated with water as the reservoir was filled. As a result of the saturation and softening effects of water, the strength of the slide mass and the sliding zone would decrease proportionally, reducing the slope’s stability. In addition, as the water level rose, the buoyant force in the lower layers decreased the slope-resistance toe’s force (Wang et al. 2008). Variations in resistance force altered the mechanical equilibrium of the mass of the slide, resulting in the failure of the slope along bedding planes of weak rocks. Thus, the Qianjiangping landslide was caused by the combined effects of Three Gorges reservoir impoundment and persistent rainfall.

3 Assessment of Reservoir Landslides Stability

Stability assessment is an important aspect in the study of reservoir landslides. In terms of the TGRA, the water level fluctuation can greatly change the hydrogeological conditions of the sliding mass. The landslides in TGRA have experienced the long-term evolution process since its formation and have tended to be stable. Therefore, although strong precipitation and earthquakes have a certain effect on the stability of landslides, a single rainstorm and earthquake will not cause the overall re-instability and failure of the landslide in normal circumstances. However, when reservoir water level fluctuates between 175 and 145 m, the generated dynamic water pressure acts on the slopes, and the softening effect of water reduces the physical-mechanical properties of the sliding mass and rupture surface, thereby reducing the stability of the slope. When evaluating the stability of landslides induced by strong precipitation and rapid drawdown of reservoir water level, the following problems need to be considered and solved:

Firstly, when taking the water table in reservoir areas into account, the concept of sudden drawdown cannot be applied to all the cases. The so-called sudden drawdown means that the decline of the groundwater level in the slope lags behind the decline of the reservoir water level, so that the excess pore water pressure generated in the slope is difficult to dissipate momentarily, and no significant amount of water is drained from the soil body during the descending process. If the permeability coefficient of the slope is large enough, the groundwater level will decrease with the reservoir water level rapidly, and the potential difference between seepage and infiltration curves will be small, in which case the sudden drawdown will not occur. At present, there is no theoretical basis for the seepage and infiltration curve formed by the reservoir water level declination, and there is no support from the measured data, resulting in too conservative landslide stability evaluation and reinforcement scheme of high slopes in the reservoir area.

Secondly, the empirical method only uses the permeability coefficient of the landslide body to estimate the stagnation height of the infiltration curve when the reservoir water level drawdown. There is a lack of research on the mechanical model of groundwater acting on the landslide, so it is difficult to fully explain the mechanism of landslides induced by reservoir water level rapid drawdown and rainstorm.

Thirdly, when calculating the pore water pressure, sometimes the lateral water pressure of the soil strip is omitted, and only the water pressure on the sliding surface at the bottom of the soil strip is calculated; Besides, some simulations are made to decide different soil weight, such as using saturated bulk density to describe the soil below the infiltration curve and above the reservoir water level, and using the floating bulk density to describe the soil below the reservoir water level, etc. Such a statement does not clearly explain the conversion relationship between the surrounding water pressure and the seepage force, which leads some people to consider both the hydrostatic pressure and the seepage force in the stability analysis, resulting in repeated calculations of the hydrostatic pressure. In fact, the penetration force is a pair of balance forces with the in the soil strip and the surrounding hydrostatic force. Considering the hydrostatic pressures as a whole is helpful to analyze the influence of seepage force on landslide stability. Therefore, how to consider the impact of the sudden drawdown of the reservoir water level on the stability of the landslide is a difficult problem encountered in the prevention and control of landslides in the reservoir area.

Based on the above understanding, according to the TGRA water level regulation scheme and the reservoir landslides groundwater action mechanics model, the transient seepage field of the landslide when the reservoir water level declines is analyzed, and the conditions for the formation of the rapid drawdown of the reservoir water level are obtained. Determination of the infiltration curve when reservoir water level drawdown and different stability evaluation methods are introduced below, and the stability of the reservoir landslides during the TGD operation are introduced subsequently.

3.1 Determination of the Infiltration Curve When Reservoir Water Level Drawdown

• (1) Saturated and unsaturated seepage models

According to the water level control scheme in the TGRA, the reservoir water level fluctuates between 175 m and 145 m. When the reservoir water level drawdown, the sliding mass above the reservoir water level is unsaturated, and the landslide mass below the reservoir water level is in a saturated state. Simultaneously, the pore water pressure gradually dissipates, and the soil changes from saturated to unsaturated state. The movement of soil water in the unsaturated area and the movement of water in the saturated area are interrelated, and the discussion unifying these two movements is the so-called as saturation-unsaturated problem (Wu et al. 2009).

When the water head h is used as the dependent variable of the governing equation, the governing equation for anisotropic two-dimensional saturated-unsaturated seepage is:

$$ \frac{\partial }{\partial x}\left({k}_x\frac{\partial h}{\partial x}\right)+\frac{\partial }{\partial y}\left({k}_y\frac{\partial h}{\partial y}\right)={m}_w{\rho}_wg\frac{\partial h}{\partial t} $$

(1)

where: k_x, k_y are the saturation permeability coefficients in the horizontal and vertical directions respectively; ρ_w is the density of water; g is the acceleration of gravity; m_w is the specific water capacity, which defined as the negative value of the partial derivative of the volumetric water content θ_w to the matric suction (u_a − u_w):

$$ {m}_w=-\frac{\partial {\theta}_w}{\partial \left({u}_a-{u}_w\right)} $$

(2)

The seepage boundary conditions are:

• ① Water head boundary

$$ k{\left.\frac{\partial h}{\partial n}\right|}_{\Gamma_1}=h\left(x,y,t\right) $$

(3)

• ② Flow boundary

$$ k{\left.\frac{\partial h}{\partial n}\right|}_{\Gamma_2}=q\left(x,y,t\right) $$

(4)

By using the two-dimensional finite element analysis program SEEP/W to simulate the steady and transient saturated and unsaturated flow, the Eq. (1) can be solved, and the seepage infiltration curve when the reservoir water level drawdown can be obtained.

• (2) Mechanical model of groundwater effects on landslides in TGRA

Compared to the seepage condition of earth dams, the seepage of landslides when the reservoir water level drawdown suddenly has more complex seepage conditions. Due to the different geological formation age of each landslide, the structural and hydrogeological conditions of the landslides are very different. Therefore, the hydrological response of each landslide is also very different when the reservoir water level rapidly drawdown. According to different hydrogeological conditions, the mechanical mechanism of groundwater acting on landslides can be summarized into the following two modes.

• ① Phreatic—aquifer mode

Pore phreatic water, founded in the alluvial-pluvial deposit, landslide accumulation layer and colluvial deposit that consists of Quaternary loose sediments, are mainly distributed in Wushan, Fengjie, Mudong, Lidu, etc. The groundwater in the landslide is mainly recharged by atmospheric precipitation and pore water. Since the permeability of rock and soil below the slip surface is relatively poor, the slip surface can be considered as confining boundary, or so-called zero flow boundary. The stability of the landslide mainly depends on the sudden drawdown of the reservoir water level. Besides, the rainstorm infiltration can increase the weight of the sliding mass above the infiltration curve and reduces the strength of the soil in sliding zone, and most of the loose accumulation landslides in the reservoir area belong to this category. In addition, if the groundwater level is buried deeply, rainstorm infiltration is difficult to raise the groundwater level.

• ② Confined water—aquifer mode

In terms of large-medium pore- or fissure-phreatic water, which is founded in the landslides such as Fanjiaping, Guling, Baota-Jipazi, Jiuxianping, etc., the bedrock is mostly disintegrated into rock blocks or even a mixture of soil and rock. The bedrock is severely weathered and broken, containing pore-fissure water or karst fissure water, and the groundwater is influenced significantly by rainfall, gullies and reservoir river water level fluctuations. At this time, the sliding surface of the landslide seepage boundary is the pressure or flow boundary. In this type of groundwater mechanics, the stability of the landslide is greatly affected by the sudden drop of groundwater level and reservoir water level. The rainstorm infiltration will raise the groundwater level, and the reservoir bank landslide is prone to occur under the effect of the sudden drawdown of the reservoir water level and the rainstorm infiltration.

3.2 Stability Evaluation Methods

Considering the variety and complexity of reservoir landslides’ characteristics in TGRA, different kinds of methods are used in the relevant stability evaluation. The methods include both qualitative and quantitative approaches, concomitantly taking the deterministic and probabilistic methods into account.

• (1) Qualitative methods

The qualitative analysis method is mainly used to analyze the various factors affecting the landslide stability, the mechanical mechanism of instability, the possible deformation and failure modes, the cause of the landslide and the evolution history, aiming at giving the quantitative description of possible trends of the evaluated landslide in terms of its stability and characteristics. The qualitative method can quickly evaluate and predict the stability of the landslide. However, the disadvantage is that this method is highly empirical and has no quantitative limit. When analyzing the landslides in TGRA, the following qualitative approaches are commonly used:

Historical cause analysis method, or so-called geological analysis method, is mainly used to deduce the whole process of the historical evolution of the landslide. This method is conducted by analyzing the geological tectonic environment (such as topography, geological conditions, etc.) and the signs of slope deformation and damage.

Graphical method is commonly used in the preliminary evaluation of landslide stability. This method generally includes stereographic projection, solid proportional projection, double-sided projection, joint pole diagram and contour diagram method, etc. If combined with the strength conditions of the structural plane and the engineering force, this method can also support the quantitative calculation of the slope stability.

Engineering geological analogy method applies the analysis experience of landslides that have been studied to the landslides with similar conditions that need to be evaluated. The landslide database, expert system and case-based reasoning evaluation method are also an extension of this method in essence. The expert system method is a computer program analysis method, which completes the analysis of landslide stability at the expert level, and combines a large amount of practical knowledge of landslide stability analysis with the problem-solving strategy of simulating human beings (Zhu et al. 2014). Moreover, studies based on database management system (DBMS) and geological information system (GIS) were employed to evaluate the reservoir areas (Bai et al. 2009). For instance, the remotely sensed graphical data can be used to produce a landslide hazard map, that indicates the relationship between the land instability and human activities (Fourniadis et al. 2007).

• (2) Limit equilibrium methods

The limit equilibrium methods investigate the equilibrium of soil mass tending to slide down under the driving and resisting forces. At present, the unbalanced thrust method (also known as the residual thrust method) is often used in the stability calculation of landslides in the TGRA, in which case the sliding mass is considered as the composition of vertical slices. In term of the stability evaluation of landslides in reservoir area, taking the seepage force generated from reservoir water level fluctuation into account is a crucial part. As shown in Fig. 15, the water force induced by the change of reservoir water level can be quantified by either the ambient pore-water pressure or by the seepage and buoyancy force (Lambe and Whitman 2018).

Fig. 15

Water force acting on a sliding mass expressed by: (a) ambient pore water pressure; (b) seepage force F_sp and buoyancy force F_b (after Zou et al. 2021)

The seepage force F_sp caused by the frictional drag of water flowing through the voids can be expressed as:

$$ {F}_{sp}={\gamma}_w iV={\gamma}_w\sin \left(\beta \right)V $$

(5)

Where γ_w is the unit weight of water, i is the hydraulic gradient, β is the slope angle of the phreatic surface, and V is the submerged volume of the analyzed soil slice, which is expressed as the area of trapezoid bcde in Fig. 15.

The buoyancy force F_b in the submerged part of the soil slice can be expressed as:

$$ {F}_b={\gamma}_wV $$

(6)

The landslide stability can be quantified by the factor of safety (Fos), which is defined as the ratio of shear resistance along the potential failure surface to the sliding force. If Mohr-Coulomb failure criterion is considered (Wang et al. 2014), Fos can be defined as:

$$ Fos=\frac{F_r}{F_s}=\frac{\sum \limits_{j=1}^n\left[c\Delta {L}_j+{N}_j\tan \varphi \right]}{\sum \limits_{j=1}^n{w}_j\sin {\theta}_j} $$

(7)

Where n is the total number of the slices, N is the normal force on the base of each slice, θ is the slope angle of the slice base (referring to the slope of sliding surface), ΔL is the length of the slice base, c is the cohesion of soil and φ is the internal friction angle of soil.

In addition to the unbalanced thrust method, the limit equilibrium methods used in the landslide stability evaluation includes Janbu method, Modified bishop method, Spencer’s method, Sarma method, Morgenstern-Price, etc. When studying the slopes and reservoir landslides TGRA, some of these methods were further modified based on specific boundary conditions and/or additional assumptions. For instance, the rock mass above the failure surface was simplified as an entire sliding block (Yong et al. 2016), and the assumptions of water level and rainfall infiltration conditions were made to proceed the equilibrium analysis (Tang et al. 2017). Besides, the influence of water level fluctuation and rainfall could be quantified in the equilibrium evaluation by introducing the time interval into the equations (Sun et al. 2016). Furthermore, the two-dimensional limited equilibrium method was extended to three-dimensional limited equilibrium method (Wang and Xu 2013).

• (3) Numerical simulation methods

Considering the complexity of stability conditions and deformation mechanism, numerical models were commonly established for the TGRA landslide cases. The finite element method (FEM) fully satisfies the statically admissible condition, strain compatibility, and the constitutive relationship between stress and strain, which is a more rigorous method for the theoretical system. Meanwhile, because the numerical analysis method can break through the limitation of the material inhomogeneity and the inconsistency of the slope geometry, it can be an ideal approach to analyzing the stress-deformation and stability conditions of landslides (Liu et al. 2007).

The recent studies use the traditional method of reducing strength index, considering the state that elastoplastic finite element calculation fails to fulfil the convergence conditions as the criterion for slope instability. Griffiths and Lane (2001) made a useful exploration in this regard. The basic principle of the slope stability analysis based on the strength reduction FE analysis is to divide the strength parameters of the slope, cohesion c^' and friction angle ϕ^', by a reduction factor F. Afterwards, new values of c^' and ϕ^' are obtained and be considered as the new soil strength parameters, which are used in the following trial calculation. When the calculation results do not converge, the corresponding F is called the minimum safety factor of the slope. At this time, the slope reaches the limit state, shear failure occurs, and the critical slip surface can be obtained.

The elastic-plastic FEM based on the strength reduction method can simulate complex geometric boundary shapes and variations of soil properties, including changes in water level and pore pressure. Generally, an ideal elastic-plastic model that satisfies the Mohr-Coulomb failure criterion is used. The elastic model is assumed to be used in the beginning of the numerical calculation, normal and shear stresses are generated at all Gaussian points within the mesh of the soil model, and these stresses are then compared to the Mohr-Coulomb failure criterion. If the stress at a particular Gaussian point locates within the Mohr-Coulomb failure circle, the point remains in the elastic state; if it is on or outside the circle, the point is yielding, and the yield stress will be redistributed in the mesh. When a sufficient number of Gaussian points yield, leading to a change of soil mechanism, the global shear failure is considered to have occurred.

When performing the FE stability analysis for landslides in TGRA, the effect from water needs to be considered. For instance, the FEM-based strength reduction method was combined with the saturated-unsaturated theory to analyze the stability of the Bazimen landslide (Zhang and Cheng 2011), in which case the effect of rainfall intensity can be evaluated. Additionally, the effects of dry-wet cycles on the sliding mass soil in a reservoir zone was investigated by conducting the hydro-mechanical couples’ simulation in the numerical modeling (Yang et al. 2022).

Furthermore, water-induced deterioration on rocks needs to be considered for the rock landslides in TGRA, since the dry-wet cycles induced by the water level fluctuation in reservoir areas reduce the rock strength and the deformation parameters. In the relevant FE analysis, the rock damage constitutive model needs to be improved by the progressive failure mechanism of rock under water-rock interaction. In addition to the rock damage evolution equation considering the initial damage, multi-parameter comprehensive evaluation of the cumulative damage effect of rock physical and mechanical properties under different water-rock interactions is conducted, and parameters based on energy theory are proposed to characterize the deterioration law of physical-mechanical properties under the action of different water and rock. Then the rock damage effect under different water-rock interaction is coupled to the damage constitutive model to perform the numerical stability analysis.

• (4) Probabilistic analysis methods considering uncertainties

The geological model, mechanical properties and external factors that are required in the stability analysis of landslides can be complex and may not be characterized with certainty. Therefore, failure probability analysis characterized by Monte Carlo simulation analysis can be applied to evaluate the landslide stability. Generally, the values of cohesion c and friction angle φ of the sliding zone are taken as random variables obeying normal distribution, and the safety factor (F) being a certain function of c and φ is also a random variable. Using Monte Carlo analysis, a series of c and φ values are produced by simulation, and then the values are applied to the limit equilibrium method to obtain the factor of safety (F). Hence the probability of F < 1 (the failure probability) is taken as the index of the stability judgment.

As a supplement to the failure probability analysis, sensitivity analysis can be used in the stability investigation of the reservoir landslides. This analysis is the study of sensitivity of a given parameter to the factor of safety by changing the one and keeping the other invariant in the stability calculation, which can ascertain the controlling factors of the landslides’ stability. Sensitivity analysis investigates the relation between the factor of safety with the different reservoir water level, ground water level, seismic intensity and sudden drawing down of the reservoir water level to flood control limitation level, and so on. Considering the factors based on the mother rock properties, the granular composition and the water content variation of the sliding zone can also be included in the analysis. Then the variation range of the shear strength parameters (c and φ) are evaluated, and the stability of landslides under various constitution of conditions can be investigated.

For examples, in the sensitivity analysis of landslides stability in TGRA, Wang et al. (2020a) considered the rainfall effects on the reservoir landslides and constructed a probability model integrating landslide susceptibility index and rainfall process when assessing the landslide hazard. In addition, Gong et al. (2020) promoted an approach of stratigraphic uncertainty modelling with random field. As explained in the paper, the spatial correlation of the stratum existence between different subsurface elements was characterized by an autocorrelation function, and the probability of the existence of a particular stratum in the non-borehole element could be determined according to the derived spatial correlations. By applying this approach to Majiazhou landslide for a probabilistic stability analysis, the influences of the stratigraphic uncertainty on the stability of this landslide and the location of the slip surface were revealed.

3.3 Stability of the Reservoir Landslides During the TGD Operation

The impoundment of the TGR altered the environmental conditions of the reservoir landslides, which had a significant impact on their stability.

• (1) Reservoir landslides stability during the trial impoundment

During the June 2003 reservoir trial impoundment, the reservoir water level rose approximately 70 m to 135 m ASL and keeps. Monitoring revealed that between 2003 and 2006, 177 landslides occurred in the TGRA after this trial impoundment. The movement of these landslides caused local cracking and destruction of houses and roads, which leads to the relocation of over 6900 people.

• (2) Reservoir landslides stability during the 156 m ASL impoundment

The water level fluctuated between 145 m and 156 m ASL during the period of the impoundment aimed at 156 m ASL, which lasted from September 2006 to September 2008. In the entire TGRA, a total of 152 landslides and significant deformations occurred, resulting in the cracking of households and roads. During this period of impoundment, approximately 3000 people were relocated, and emergency investigation, monitoring, and rescue operations were conducted with approximately 180 emergency disposals in the reservoir area.

• (3) Reservoir landslides stability during the 175 m ASL impoundment

The 175 m ASL impoundment of TGRA was conducted after the 2008 rainy season. The impoundment of the reservoir began on September, 2008, and ceased on October 8, 2008, when the water level reached 156 m ASL. Then, the reservoir impoundment restarted on October 17, and the water level reached 172 m ASL on November 14. After that, the water level of the reservoir slowly declined. By the end of March 2009, the water level had fallen back to 159 m ASL, then to 155 m ASL at the end of May and to 145 m ASL on 10 June 2009. During this period, 196 landslides and reservoir bank deformation events occurred in the TGRA. The former included the November 2008 North Nierwan landslide in Zigui County, the April 2009 Liangshuijing landslide in Yunyang County and the October 2009 Taping landslides in Wushan. Overall, the landslides and bank deformation caused the cracking of buildings and roadways on the reservoir banks, threated 24,000 residents, and posed a threat to Yangtze River shipping.

4 In-Situ Experiment on Slipping Zone Soil of Reservoir Landslides

4.1 In-Situ Direct Shear Creep Test

As the shear zone in BT3 is much thicker than that in BT5 and the hydrological condition in BT3 is more suitable for the in situ test, the testing tunnel in BT3 was selected as the in situ creep test site, as shown in Fig. 16b. Several samples were prepared in the east direction of the testing tunnel in BT3. To this end, the left side of the testing tunnel was excavated, leaving a platform with shear-zone soil above the shear surface. The shear-zone soil was then prepared to block samples for the in situ creep tests. To mimic the landslide movement as realistically as possible, in situ creep shearing was conducted parallel to the shear surface, as shown in Fig. 16c. A direct shear apparatus was designed for the purpose of long-term reliable in situ creep testing at the field site.

Fig. 16

(a) The tunnels in the Huangtupo landslide, (b) creep test area in the testing tunnel in BT3, and (c) the direct shear device fixed on the shear surface (after Wang et al. 2020b)

Through the in situ direct shear creep test carried out at the Badong field test site, the creep properties of the clastic soil within the shear zone were examined over various shear stress ranges. Additionally, some laboratory ring shear creep tests were carried out with the same soil. With the comparison of the test results and those obtained by triaxial creep tests reported in the literature, the following discoveries were drawn: The in situ direct shear creep test yields a failure stress of approximately 0.95 of the peak shear strength. Compared to the ring shear creep test, a much larger rate of creep displacement was obtained due to shearing along the landslide direction. The tested sample in the in situ direct shear creep test exhibits very low compressibility during creep due to the high content of coarse particles (25 and 50 wt%) and over consolidation. In the ring shear creep tests, the critical creep stress was approximately 0.98 of the peak shear strength at the prepeak state. However, at the residual state, the critical creep stress was equal to the residual strength of the soil. The shear-zone soil is prone to failure under shear conditions rather than triaxial compression conditions. Nonattenuating creep responses were obtained in both the in situ direct shear and the laboratory ring-shear creep tests, while no accelerated failure was observed in either the laboratory or the in situ triaxial creep tests.

4.2 In-Situ Triaxial Creep Test

To investigate the characteristics of the in situ sliding zone soil, which is distinguished by a high rock content and high degree of compression, an in situ triaxial creep test was performed in BT3 as well. A subdivision of the Huangtupo landslide called Riverside Slump #1 is where the examined sample was found (Fig. 17). The test took 191 days to complete, after which the results were examined.

Fig. 17

In-situ triaxial creep test on sliding zone soil of the Huangtupo landslide (after Tan et al. 2018b)

The test results showed that only decaying creep occurs without abrupt failure. A constitutive equation with both linear and nonlinear viscoplastic terms was deduced to accurately fit the test data. The surface with an orientation of 35° presents anisotropic traits in terms of displacement, possibly due to cracks that have formed over time; and the sliding zone exhibits low compressibility due to the high rock content (54.3%) and long-time consolidation by the overlying soil mass. The sliding zone soil behavior resembles the behavior of a landslide when a stress magnitude similar to that of the in situ stress environment is applied to the sample. The deformation and characteristics of the sample, taken together with the test results, suggest that the Huangtupo 1# landslide will continue to creep.

5 Prevention and Control of Reservoir Landslides

Due to the non-negligible effect of water, the most important measures to prevent or control reservoir landslides is drainage, combined with slope reprofiling and unloading, construction of stabilizing piles, prestressed anchor cables, anchors, lattice beams shore protection works, dynamic monitoring and composite retaining structures, etc. (Zhang et al. 2014b). The prevention and control of reservoir landslide should be based on its type, scale, stability, and combined with the regional engineering geological conditions, hazard objects and construction season and other conditions, and the comprehensive use of a variety of forementioned management forms. Studies in aspect of stabilizing piles, anchoring, and practical treatment projects of reservoir landslides are described as below.

5.1 Stabilizing Piles

In reservoir areas, stabilizing piles was one of the most common and effective methods for preventing and mitigating landslides. Numerous studies have focused on the theory and analysis of stabilizing piles in the past decades. The landslide thrust, interaction mechanism between stabilizing piles and landslides, and design optimization were investigated using numerical simulation techniques, laboratory, and in-situ tests (Dai 2002; Tang et al. 2016). In the landslide thrust analysis, the landslide evolution process that accounts for water fluctuations was also considered (Jia et al. 2009; Tang et al. 2014).

The interaction mechanisms between the stabilizing pile and landslide include those between the stabilizing pile and sliding mass (pile–soil interaction) and those between the stabilizing pile and bedrock (pile–rock interaction). Both are essential for the optimal design of stabilizing piles. In the study of pile-soil interaction mechanism, the effect of soil arching was considered (Tang et al. 2014; Zhang et al. 2018). In terms of pile-rock interactions, physical model tests and numerical studies were frequently utilized. Li et al. (2016) studied the behavior of stabilizing piles with different embedded lengths in bedrock with a strong upper layer and a weak lower layer using physical model tests.

Design optimization is a crucial component of stabilizing pile design. Theoretical and numerical methods based on soil arching effects were frequently employed to optimize the cross-section size and piles’ spacing (Li et al. 2016; Wu et al. 2017). A deformation control principle was developed based on the multilayer subgrade coefficient method and physical model tests to determine the optimal pile length (Li et al. 2019). Theoretical and numerical methods that consider the safety requirement, the critical slip surface, and the stress and deformation of the stabilizing piles were used to determine the optimal pile location (Li et al. 2017).

5.2 Anchors

Due to the advantages of low cost, high efficiency and easy operation, anchors have been widely used in TGRA landslides. However, theoretical research on anchorages has been lag behind in some aspects already. The engineering analogy method and the semitheoretical and semiempirical methods are still commonly used in engineering practice (Liu et al. 2018). The design of the anchoring system requires constant optimization.

The study of interaction mechanism between anchor and rock mass (soil) is of great significance. The pullout characteristics of anchorage, which include the stress transfer between the anchor, mortar and surrounding rock, should be the most critical point. The variety of strata leads to complicated stress distribution. For instance, considering the nonuniform stress of the horizontally layered rock mass along anchors, the load transfer characteristics of anchors in the rock mass with weak interlayers were investigated. It was found that due to the existence of weak interlayers, the shear stress mutates at the weak and hard rock interface, and the axial force appears to “rebound” at the bottom of the anchored section (Wang et al. 2021b). Song et al. (2022) revealed that the shear stress distribution and the axial stress distribution were characterized by negative exponential attenuation, and the time-history variation law of the prestress was explored. The prediction formula was verified in Maya High Slope at the Shuibuya Hydropower station.

In addition, as the anchors pass through the sliding zone, they are subjected to a strong shearing when there is relative displacement between the sliding mass and sliding bed; hence, there is a growing body of literature that recognises the importance of anchor shearing. In laboratory tests, the stress of external environment (Zheng et al. 2021a), the properties of rock masses and discontinuities, for instance, JRC of rock joints (Wang et al. 2022), and the anchorage parameters and modes (Zheng et al. 2021b), can be significant factors for the effectiveness of anchor support and the stability of landslides. In theoretical aspect, considering the relationship between the dilation effect and the bolt deformation and the influence of joint surface abrasion on the attenuation of the internal friction angle, Zheng et al. (2021c) proposed an analytical model for predicting the shear behavior of bolted rough joints, which successfully predicted the shear load-shear displacement curve of the anchoring joint and the change curve of the anchor shaft force.

In terms of the design optimization of anchor, the direction angle should be of particular importance. An et al. (2020) proposed a new method to calculate the optimal anchorage direction angle, for case of the reinforcement direction perpendicular to the strike direction of the slope, derivative method was used, and when the reinforcement direction has no restrictions, the fmincon function was adopted to optimize the anchorage direction angle. The proposed method can further improve the anchoring effects and reduce the total usage of slope supporting costs.

The reservoir landslides may lead to anchor shearing (Fig. 18). However, modeling the shear mechanical behaviors of bolted rock joints is very complicated due to the complex factors that affect the axial and shear forces on the bolts. In the process of shearing, the rough surface of the joint undergoes a dilation effect and morphological abrasion, which is very common in reservoir landslides, were highlighted in our model (Zheng et al. 2021b).

Fig. 18

Schematic diagram of anchor under shearing

The key steps of the model are as follows: the contribution of the bolt to the shear strength is a combination of the axial and shear forces (Fig. 19), based on the Mohr-Coulomb criterion, the following formula can be applied to compute the shear loads of the bolted rock joints:

Fig. 19

Force of the bolt under shear action

$$ {T}_{bj}=\left({R}_c+\Delta {R}_{c_b}\right)+\left(\Delta {R}_{\sigma_{nb}}+{R}_{\sigma_n}\right)\tan {\phi}_j $$

(8)

$$ \Delta {R}_{c_b}=\left({N}_{\mathrm{O}}\sin {\omega}_{\mathrm{O}}+{Q}_{\mathrm{O}}\cos {\omega}_{\mathrm{O}}\right) $$

(9)

$$ \Delta {R}_{\sigma_{nb}}=\left({N}_{\mathrm{O}}\cos {\omega}_{\mathrm{O}}-{Q}_{\mathrm{O}}\sin {\omega}_{\mathrm{O}}\right) $$

(10)

Where T_bj is the shear load of the bolted rock joint; R_c is the shear load provided by the cohesion of the joint surface; $\Delta {R}_{c_b}$ is the shear load provided by the additional cohesion of the bolt; $\Delta {R}_{\sigma_{nb}}$is the additional normal load provided by the bolt on the joint surface; ${R}_{\sigma_n}$ is the initial normal load of the joint surface; ϕ_j is the joint friction angle; and ω_o is the rotation angle of Point O.

The internal friction angle is not a constant (Fig. 20), and the evolution process of the internal friction angle can be described by following functions:

$$ {\phi}_{\mathrm{e}}=\left\{\begin{array}{l}{H}_1{\delta_v}^2+{H}_2{\delta}_v+{H}_3\kern2.25em {\delta}_v\in \left[0,{\delta}_d\right)\kern0.5em \\ {}{H}_4{\delta_v}^2+{H}_5{\delta}_v+{H}_6\kern2.25em {\delta}_v\in \left[{\delta}_d,{\delta}_e\right)\ \\ {}{H}_7\kern8.75em {\delta}_v>{\delta}_e\end{array}\right. $$

(11)

where H₁, H₂, H₃, H₄, H₅, H₆ and H₇ are undetermined coefficients and δ_v is the shear displacement.

Fig. 20

Calling process of the friction angle with shear displacement

In this model, a bolt is regarded as a semi-infinite beam on the elastic foundation, and the elastic foundation beam theory is used to calculate the shear force:

$$ {Q}_{\mathrm{O}}=\left\{\begin{array}{l}\frac{110{\sigma_{\mathrm{c}}}^{0.85}{v}_{\mathrm{o}}}{\lambda}\kern4.75em , DI\le 0.02\\ {}\frac{110{\sigma_{\mathrm{c}}}^{0.85}{v}_{\mathrm{o}}}{\lambda {\left(1+3{\left(\frac{v_{\mathrm{o}}}{D_{\mathrm{b}}}-0.02\right)}^{0.8}\right)}^4}, DI>0.02\end{array}\right. $$

(12)

where σ_c is the compressive strength of grouting concrete; v_o is the transversal displacement of the bolt (i.e., bolt deflection); D_b is the bolt diameter; λ is bedding coefficient; DI is a dimensionless damage coefficient.

Considering the dilation effect and the influence of normal stress (Fig. 21), the calculation formula of the axial force of the bolt at joint is shown in function (13), Where ε_L is the strain of AO; A_b is the cross-sectional area of the bolt; ε_e is the strain corresponding to the yield strength of the material; k is the modulus of the elastic foundation (i.e., spring stiffness); E is the elastic modulus, E_p is the plastic modulus; ω_op is the is the plastic rotation angle of Point O.

$$ {N}_{\mathrm{C}}=\Big\{{\displaystyle \begin{array}{l}{\varepsilon}_L{A}_bE\cos \left({\omega}_o\right)-\frac{kv_{\mathrm{O}}}{2\lambda}\sin \left({\omega}_o\right)+\frac{3{kv}_{\mathrm{O}}}{4\lambda}\sin \left(\frac{11}{18}{\omega}_o\right){\varepsilon}_L<{\varepsilon}_e\\ {}\left({\varepsilon}_e{A}_bE+\left({\varepsilon}_L-{\varepsilon}_e\right){A}_b{E}_p\right)\cos \left({\omega}_{op}\right)-\frac{kv_{\mathrm{O}}}{2\lambda}\sin \left({\omega}_{op}\right)+\frac{3{kv}_{\mathrm{O}}}{4\lambda}\sin \left(\frac{11}{18}{\omega}_{op}\right){\varepsilon}_L>{\varepsilon}_e\end{array}} $$

(13)

The tensile-shear failure mode of the bolt adopts the Tresca strength criterion:

$$ {\left(\frac{N_o}{N_p}\right)}^2+{\left(\frac{Q_o}{Q_p}\right)}^2=1 $$

(14)

where N_p and Q_p represent the corresponding ultimate axial force and shear force, respectively, under noninteractive force conditions; N_p = A_bσ_p; Q_p = A_bσ_p/2; and σ_p is the ultimate axial stress at failure.

Fig. 21

Shear deformation process of the bolt

Through the above steps, the shear load-shear displacement curves of bolted rock joints can be obtained, we verified the model by conducting laboratory tests. Figure 22 shows the analysis models of the roughness of five different joint surfaces and the results of the direct shear tests. The analysis model shows agreement with the experimental results, which verified the validity of the model. The model reveals that it is necessary to consider the dilation effect and morphological abrasion of rock joint when the bolt is subjected to shear action in reservoir landslides.

Fig. 22

Comparison between the calculated value and experimental value: (a) JRC = 0; (b) JRC = 3.14; (c) JRC = 6.99; (d) JRC = 12.01; and (e) JRC = 16.07

5.3 Case Studies

To examine the effectiveness of the proposed method for the determination of stabilizing piles location in the reservoir landslides with interbedding of hard and soft bedrock, the Yanguan landslide located in Yichang city, China, was taken as an example (Fig. 23). With the proposed method, it was found that the location of the hard layer in bedrock has a significant effect on the pile deformation. The impact of hard bedrock layer on piles relies on its location coefficient. The optimal pile location corresponds to a region with both hard top and bottom layers in bedrock. The optimal location of stabilizing piles should be located in the hard top and bottom bedrock layers, with a considerable decrease in both the displacement and internal force of stabilizing piles.

Fig. 23

Engineering geology plane of the Yanguan landslide (after Li et al. 2017)

Fig. 24

(a) Drainage system used to stabilize the Huanglashi landslide, whose toe is affected by TGR water level fluctuations (note the 145 m and 175 m contours); (b) Full view of the Huanglashi landslide; (c) Aerial photo showing surface drainage system

Drainage, a straightforward but frequently effective method for preventing landslides, was frequently utilized in some larger and deeper reservoir landslides with significant movement. The Huanglashi landslide in Badong County is a successful example of hazard control by drainage. The volume of the landslide is 18 million m³, and its hydrogeological and engineering geological conditions are complex (Guo et al. 1999). The variation of groundwater was found to have a significant effect on the stability of the Huanglashi landslide. Thus, a drainage system was implemented to stabilize and prevent further landslide development (Fig. 24). The following control schemes were implemented: On the central portion of the Huanglashi landslide, a row of evenly spaced wells was installed initially. It contained 33 drainage wells of 20–30 m deep with distances ranging from 5 to 10 m between them. Then, an adit was installed to collect water beneath the sliding zone. Water entered the adit via drainage wells and then flowed into surface ditches. Based on the monitoring and observations (giving dates or periods), it was found that the implemented control measures were quite effective.

6 Forecasting of Reservoir Landslides

6.1 Forecast Model of Reservoir Landslides

In the field of geohazard prevention and control, reservoir landslides forecasting, especially temporal forecasting, is a long-standing challenge that has received considerable attention. The primary objective of reservoir landslides short-term forecasting is to determine the critical sliding time, whereas long-term forecasting focuses on predicting long-term deformation based on displacement-time series under the influence of multiple external and internal factors. The forecast methods can be classified into four categories: physical-mechanical forecast methods, traditional nonintelligent methods, intelligent methods, and probabilistic forecast methods.

Since the proposal of the Satio model, the physical-mechanical forecast model was established based on the analysis of large-scale laboratory experiments and field monitoring data, which are suitable for short-term forecasting. In consideration of the complexity, time consumption, and cost of experiments, physical-mechanical forecast models have strict research and application restrictions, and research has gradually slowed down in recent decades. Based on creep theory, these models describe the deformation of landslides mainly in the forms of rheological functions. In the reservoir landslides long-term forecasting, traditional nonintelligent methods, intelligent methods, and statistical forecasting methods are primarily used nowadays. For traditional non-intelligent methods, prediction models are often developed from algorithms such as Wavelet transform, grey theory, time series, Kalman filter, etc. to forecast the behavior of reservoir landslides in the TGRA (Xu et al. 2011). As the artificial intelligence algorithms rapid developed recently, intelligent methods such as artificial neural network (ANN) and supporting vector machine (SVM) have become more prevalent in the long-term prediction of reservoir landslides (Bai et al. 2009; Du et al. 2013; Ma et al. 2017). Intelligent methods are effective at handling complex problems associated with landslide forecasting, and the effect of uncertainty in these methods has raises widely attentions as probabilistic models that can explicitly consider the uncertainty have been developed in the recent years (Ma et al. 2018).

In order to mitigate the reservoir landslides effectively, an early warning criterion that issues timely alarms is also vital.

Typically, the early-warning criteria for reservoir landslides in TGRA are based on macroscopic precursors, stability indices, displacement data, and environmental conditions. Examples of macroscopic precursors include the significant deformation of the ground surface and buildings, the collapse of the landslide toe, and the sudden variation of subsurface water table. During the critical sliding stage, these phenomena are distinct and generally detectable by native residents. The stability indices are based on the of safety factor and the reliability index, which are derived from deterministic and probabilistic analyses, respectively (Hu et al. 1996). Displacement data represent the most direct and accessible information obtained through landslide monitoring. Consequently, many criteria, such as displacement rate and acceleration (Yuan et al. 2015) and displacement vector angle, have been proposed based on displacement data (He et al. 2003). Criteria based on the environmental conditions are mainly related to rainfall and TGR water level fluctuation. Maximum 24-h rainfall intensity and reservoir water level decline rate are effective indicators for early warning of TGRA landslides (Ni et al. 2013).

Except for the physical-mechanical forecast model, universal and accurate prediction of the aforementioned methods is a bottleneck problem, as they concentrate on the characteristics of landslide behavior and external dynamic factors. To overcome this difficulty, the development of a reservoir landslide forecast model based on its physical-mechanical mechanism should become a priority research topic.

6.2 Framework of Physical-Mechanical Forecast Model

To develop the reservoir landslides prediction model based on physical-mechanical mechanism, works has been done in three aspects: (1) development of comprehensive multi-field spatiotemporal correlation monitoring system; (2) development of multi-source monitoring data fusion method; (3) numerical prediction model. The fusion of the above studies can finally realize the construction of the reservoir landslides real-time prediction platform based on the physical-mechanical model.

• (1) Comprehensive multi-field spatiotemporal correlation monitoring system

The first step in developing physical-mechanical forecast model for reservoir landslides is to build a 3D comprehensive monitoring system of reservoir landslides that includes multi-field spatiotemporal correlated monitoring technology. The developed monitoring system allows for the real-time and effective acquisition of key parameters of prediction geological criteria and initiation criterion. The Huangtupo landslide in the TGRA has been chosen, and a 3D comprehensive monitoring system was built on it (Fig. 25).

Fig. 25

3D comprehensive monitoring system in Huangtupo landslide

At various latitudes, surface displacement and meteorological data were monitored by space-air-ground technology. Through interferometric synthetic aperture radar (InSAR) and multi/hyperspectral remote sensing satellites, the landform and displacement distribution of the entire landslide and its surrounding area was obtained, and the sensitive area of landslide dynamic evolution can be quickly identified. Distributed monitoring of local sensitive areas of landslide was carried out by using technologies and methods such as drone laser scanner (DLS), unmanned aerial vehicle (UAV), ground-based InSAR, etc., from which only a few representative feature sites were selected for further monitoring. Global navigation satellite system (GNSS), borehole monitoring, and other methods were used to monitor these representative feature sites to obtain aging effective characteristic parameters that accurately reflect the landslide evolution state. To achieve continuous monitoring of reservoir landslide multi-field under large deformation, a multi-point distributed flexible inclinometer and deformation coupled pipeline trajectory inertial measurement were developed. On this basis, a multi-field spatiotemporal correlated monitoring technology integrating deep displacement, groundwater level, earth pressure was developed. The obtained monitoring parameters of various types have a high degree of spatiotemporal consistency and correlation, which can effectively improve data information density and prediction value.

• (2) Multi-source monitoring data fusion of reservoir landslides

Aiming at the defects that the fusion of reservoir landslides multi-source monitoring data and their evolution process and physical-mechanical mechanism are disconnected, the structural form and features of landslide multi-source monitoring data were analyzed first (Fig. 26). Then, the classification and coding specification of multi-source monitoring data of landslide was established, and the standardized processing method of multi-source heterogeneous monitoring data was proposed to realize the establishment of a priori model of nonlinear system of landslide evolution process. Study on the time-series correlation of multi-source monitoring information during landslide evolution process is carried out to reveal the self-consistent rules between the nonlinear priori model of landslide and multi-source monitoring information. A time-domain fusion method of landslide multi-source monitoring data driven by a mixture of geological processes and data was proposed so that the key field characteristic attributes of landslide evolution and the logical co-occurrence structure of multiple fields can be extracted. After the sequence pattern of landslide multi-field correlation monitoring data was identified, the critical characteristics of evolution state transition was mined and the landslide evolution state transition equation was established, the geological mathematical analysis method of landslide multi-source monitoring information can be developed eventually.

Fig. 26

Multi-source monitoring data fusion technology

• (3) Numerical prediction model of reservoir landslides

Based on the initiation mechanism and mechanical model of reservoir landslides, a generalized intrinsic evolution model of reservoir landslides was developed by characterizing the mechanical process under the constraints of the dominating structures of geo-material, as well as complex and volatile hydraulic forces. Then, the physical-mechanical evolution model of reservoir landslide was established based on the multimap between the spatial-temporal varying pivotal characteristic parameters of the reservoir landslide (e.g., displacement, velocity, acceleration, stress, pore pressure, etc.) and the constitutional model of the critical geo-material (e.g., slip zone, weak intercalation in the rock mass, soil-bedrock interface, etc.). Meanwhile, the probabilistic characterization and modeling methodology of the field-scale geological model and external dynamic factors in terms of fluctuating reservoir water level and rainfall were proposed, and a probabilistic prediction model of landslide initiation based on physical-mechanical process was established. It enables real-time ensemble modeling of the landslide evolution process based on variable geological elements, initial conditions, and external hydraulic forces. These variable factors can be inversed dynamically based on Bayesian theory and multi-field monitoring data in real time. The updated factors was employed to predict the forward evolution behavior of landslides, thus enabling real-time inversion and dynamic updating. Therefore, a numerical landslide forecasting (NLF) mode was established by integrating probabilistic characterization of the geological model and external dynamic forces, probabilistic forecasting of landslide initiation, and real-time dynamic update of landslide evolution processes, as shown in Fig. 27.

Fig. 27

Schema of reservoir landslides numerical prediction

6.3 Real-Time Prediction Platform

NLF is a radically different paradigm for landslide prediction with two aspects of requirements. First, the physical-mechanical model, the core of the mode, should be as sophisticated as possible. It is required not only to identify the progressive evolution of landslides based on the evolution phase, but also to characterize the evolution process employing the stress-strain-strength-time correlation of the key geo-material corresponding to the actual stress traits to the greatest extent possible. Moreover, the multi-field correlated monitoring of landslides should capture the variations of the most critical evolutionary parameters as much as possible.

On this basis, establishing a real-time prediction platform is an effective way to achieve NLF. Comparing the current geohazard and meteorological real-time prediction platforms, it is proven feasible to build a clouding platform. The main functions of the platform include four parts: storage and recall of background data of landslide, real-time acquisition and transmission of multi-field monitoring data, real-time pre-processing of multi-source data, numerical forecast and further information distribution, as shown in Fig. 28.

Fig. 28

Schema of reservoir landslides real-time prediction platform

The landslide background data system covers a series of basic geological data, including landslide formative background, fundamental geological data, engineering geological data, hydrogeological data, remote sensing data, etc. Based on the research and development of multi-field correlation monitoring technology, a space-air-ground-underground integrated monitoring network can be constructed through the Internet of Things (IoT) and other information transmission technologies to obtain landslide monitoring data in real time. The multi-source monitoring data fusion technology can be employed to implement clouding data pre-processing and computing, thus synchronizing the display of monitoring data on the client side and the cloud. Subsequently, the ensemble modeling of the landslide evolution process can be implemented based on uncertainty theory with the geological elements, initial and boundary conditions as variable inputs. Bayesian networks and Markov theory are then adopted to calibrate and update the physical-mechanical model and inputs with multi-field monitoring data in real time, and derivate the entire evolution process of landslide and implement dynamic prediction based on the updated boundary conditions and parameters in the cloud. It is similar, but not identical, to data assimilation. Data assimilation is essentially the process of incorporating new observations into the dynamic operation of a numerical model to bring the predicted values closer to the true values. In contrast, the process of dynamic inversion in NFL is essentially a process of correcting the inputs of the model with observations to achieve dynamic prediction. The computed landslide process and probabilistic forecasting information of landslide initiation are then submitted to the PC terminal for backup storage and dynamic display. Depending on the initiation criteria and threshold, the landslide evolution and initiation probabilistic forecasting information can be further distributed and alerted to particular users.

7 Challenges and Suggestions

The reservoir landslide research conducted over the past several decades has generated a vast amount of valuable scientific and technical literature. These works have enhanced our comprehension and capacity to prevent and mitigate their effects. Although significant achievements have been made, numerous challenges remain. To overcome these challenges, future research must invest significantly more time and resources. The following list of challenges and suggested topics is intended to stimulate further research.

• (1) 3D fine geological modeling of reservoir landslides

The macroscopic geomechanical model of reservoir landslides is vital in the assessment of reservoir landslides and their evolution mechanisms. The consideration of the combined action of multiple fields, phases and factors needs to be further deepened in the traditional reservoir landslide geomechanical modeling method. Modern noncontact investigation techniques, such as small unmanned aerial vehicles (UAV), remote sensing, 3D laser scanners and LiDAR, should be adopted and systematically integrated in reservoir landslide field investigations. Combined with the detailed field investigation, the refined modeling of the three-dimensional geological model of landslides in complex strata in the reservoir area can be realized.

• (2) Long-term stability assessment of reservoir landslides

Prior research concentrated primarily on the rheology and residual strength properties of the sliding mass and the slip surface. Nevertheless, the reservoir landslides involve a process of dynamic stress change and deformation caused by external factors. These dynamic changes must be taken into account and explicitly incorporated into the long-term stability assessment of reservoir landslides. More research is required to evaluate the long-term stability of reservoir landslides stabilized by engineering structures.

• (3) Construction of in-situ experimental station of reservoir landslides

The comprehensive platform for landslide in situ testing is an important means to study the formation and evolution mechanism of reservoir landslides. By constructing the field in situ test station of reservoir landslides and carrying out the in situ mechanical test and comprehensive monitoring demonstration project, valuable first-hand test and monitoring data can be obtained to further confirm the results of theoretical and numerical analyses, which is conducive to promoting research on the long-term evolution evaluation of reservoir landslides under complex conditions (such as reservoir water fluctuation and rainfall conditions).

• (4) Multi-parameter integrated monitoring system for reservoir landslides

The evolution process of reservoir landslides can be more accurately revealed through the integrated acquisition of multi-parameter data that may be cross-correlated. However, the current monitoring systems was usually in the form of the randomly combination of different monitoring devices, which collect the data independently. Thus, an integrated monitoring system of multi-parameter is urgently needed. The integrated and correlated information regarding the evolution of various fields inside reservoir landslides, such as reservoir seepage, stress, and surface/subsurface displacement, provides a more solid foundation for more accurate landslides analysis. Research of integrated monitoring technology, as well as new algorithms for effectively utilize the multi-parameter correlated information in the rapid assessment of landslide stability and deformation is required (Tang et al. 2019).

• (5) Control countermeasures for huge reservoir landslides

In TGRA, over 30 enormous landslides (volume >10 million m³) have been identified. Many of these landslides, such as the Huangtupo and Baishuihe landslides, are still active and pose a significant threat to Yangtze River navigation with cumulative displacements of several meters per year. Due to the size of these landslides, however, traditional control methods such as anti-slide piles and anchor cables are difficult to implement. The creation of efficient preventative measures to manage and control these massive landslides is urgently needed.

• (6) Combination of prediction model and physical-mechanical mechanisms for reservoir landslides

The numerical models used to predict reservoir landslides are based on mathematical methods and statistical theories that are unrelated to the landslide evolution process and physical-mechanical mechanism. These models’ forecasting universality and success rate have always been an unavoidable issue. Thus, if forecasting models are not fully integrated with reservoir landslide physical-mechanical mechanisms, the theoretical and technical bottlenecks of landslide prediction and forecasting will never be broken.

8 Conclusions

The study of reservoir landslides has been a major focus of the engineering geology research, and received special attention in the construction of the TGD in China in the past few decades. These efforts have led to better knowledge and practice in the identification and mitigation of reservoir landslides. A comprehensive introduction of the reservoir landslides related articles and research developments in the TGRA are presented here. First, the studies of characteristics of reservoir landslide in TGRA in aspect of spatiotemporal distribution, slide-prone strata, influence of geologic structure, dynamic characteristics, influence of rainfall and fluctuating water level was summarized. Typical seepage-induced reservoir landslide and buoyancy-induced reservoir landslide from TGRA are introduced and analyzed, respectively. Then, the assessment methods for reservoir landslides in the past few decades was examined. After that, large-scale in-situ experiments, including in-situ direct shear creep test and in-situ triaxial creep test, conducted inside the Huangtupo landslide was described, which offered a unique research opportunity for a better understand of sliding zone soil’s properties. Hereafter, research progress in aspect of stabilizing piles, anchoring, and practical treatment projects were described. Finally, the prospects of reservoir landslide forecasting based on physical-mechanical mechanisms were elaborated.

Despite the major advances in the research and development, challenges still exist in the identification and mitigation of reservoir landslides. These include issues such as the 3D fine geological modeling of reservoir landslides, long-term stability assessment of reservoir landslides, construction of in-situ experimental station for reservoir landslides, multi-parameter integrated monitoring system for reservoir landslides, control countermeasures for huge reservoir landslides and the combination of prediction model and physical mechanics mechanism for reservoir landslides. These challenges represent real opportunities for the engineering geologists in their pursuit of reservoir landslides-related hazards reduction and mitigation.