Physical Mechanism and Numerical Simulation of Landslide Dam Formation

Abstract

Landslide dam formation creates one of the most hazardous geomorphic phenomena, which often threaten the safety of upstream and downstream communities by causing various secondary hazards, such as backwater inundation, dam breaches, debris flows, and outburst floods. Understanding the conditions and processes of the landslide dam formation is useful for hazard assessment and its disaster countermeasures. This research presents the physical mechanism of landslide dams and their entire formation process through three typical cases, namely the large-scale Kuridaira and Akatani landslide dams in Nara prefecture, Japan and the massive Jure landslide dam in Sindhupalchok district, Nepal, by using ring shear tests and integrated simulation with LS-RAPID model. The results from ring shear tests show that the samples of the sliding surfaces underwent substantial grain crushing and sliding-surface liquefaction due to a rapid generation of pore water pressure and significant loss of shear strength. In the computer model, the formation processes of the three landslide dams consisted of four main stages from stable slopes to local failures through progressive failures to massive slope movement at high speeds and river damming at the end of the motion. The analysis indicates the high mobility behavior that governs the rapid motion of the landslides is a primary contributing factor to the dam formation.

1 Introduction

Landslide dams are one of the typical geomorphic processes whereby the downslope movement blocks a river channel, forming a reservoir in the valley floor. Most landslide dams are formed by rock/debris avalanches, rock/soil slumps and slides, mud/debris/earth flows triggered by excessive rainfall and snow melt, and earthquakes (Schuster and Costa 1987). Landslide dams can be classified into six types based on geomorphological features (Schuster and Costa 1987) or three classes based on their evolution, including unformed dams, formed-unstable dams and formed-stable dams (Stefanelli et al. 2015). The formed-stable dams form a lake basin due to a complete stream blockage. In comparison, the formed-unstable type indicates the complete river damming to create a lake basin. It remains for a variable period (from hours to centuries) until external contributing factors can trigger the failures. Although the complete damming can persist for varying durations from few minutes to many thousands of years, many dam reservoirs tend to fail within 1 year from their initial formation (Korup and Wang 2015). The landslide dam evolution often poses sequential hazards to the public in upstream and downstream regions. The potential hazards resulting from the dam formation are upstream inundation, overflow and overtopping, dam breach, outburst discharge, downstream flooding, debris flows, and river bank erosion (Schuster and Costa 1987; Korup 2005). There are numerous recorded events of landslide dams that have caused disastrous effects on communities; for example, the 2009 Shiaolin landslide dam caused over 400 fatalities in Taiwan (Tsou et al. 2011), and the 2014 Jure landslide dam claimed 156 lives in Nepal (Tien et al. 2021). Notably, the 2013 catastrophic Kedarnath flood disaster resulting from the breach of the Chorabari Tal Glacial landslide dam killed more than 5000 fatalities in Uttarakhand, India (Ray et al. 2016). It is, thus, essential to assess the possibility of landslide dam formation and associated secondary hazards for disaster risk reduction and prevention.

Many researchers have investigated the mechanisms and conditions governing landslide dam formation. Among them, the geomorphic approach is widely used based on global and regional datasets of landslide dams (Costa and Schuster 1987; Korup 2002, 2005; Ermini and Casagli 2003; Fan et al. 2012). The topography of narrow valleys surrounded by steep slopes and high mountains is the preliminary setting for the evolution of landslide dams. This feature is commonly in regions with active geological processes, where slope materials are characterized by fractured and sheared or hydrothermally altered bedrock. These topographical and geological features make slopes very susceptible to landslides. Narrow valleys are crucial factors that increase the possibility of complete channel blockage. It facilitates the potential of river-damming completion by relatively small volumes of sliding. The high and steep terrain is often prone to rapid-moving landslides that can quickly block rivers before their deposit can be swept away by water flows (Schuster and Costa 1987). Notably, several geomorphological indexes that combine various morphological parameters related to the landslides (such as: the volume, depth, and travelling speed) and the river valley (e.g., slope parameters, catchment area, valley width) have been extensively suggested to evaluate the landslide dam evolution for its risk assessment (Costa and Schuster 1987; Korup 2004; Evans 2006; Ermini et al. 2006; Stefanelli et al. 2017).

Landslide dams have complex geomorphic characteristics and cascading processes (Dal Sasso et al. 2014). Although the conditions for the dam formation are strongly governed by the initiation and motion of the landslides, the physical mechanisms of their formation have not been fully understood so far (Tien 2018). Literature reviews show that research topics on landslide dams are primarily descriptive and mainly focused on susceptibility assessment of dam failures associated with secondary hazards, e.g. dam-break floods and backwater inundation. Landslide dam formations were separately addressed as a single process consisting of the occurrence of landslide itself, the creation of dams and its natural lakes, or the potential of secondary hazards (Korup 2002). Since the hydrodynamic interaction between landslides and rivers, and the involved processes for dam formation have yet to be established in an approved theory, modelling the entire process of dam evolution is still developing (Dal Sasso et al. 2014). The simulation of landslide dam formation with input parameters has only based on the trial-and-error analysis, calibration-based analysis or back-analysis instead of using soil physical properties. These previous 2D models could only investigate the kinematic effects of the landslides and its interaction with the terrain of slopes. However, they could not reproduce the effects of groundwater changes and pore water pressure build-up due to rainfall, Tien et al. 2021). Therefore, this study aims to physically investigate the formation mechanisms and simulate the landslide dam formation through an integrated landslide simulation model of the ring shear apparatus (ICL-2) and the computer model (LS-RAPID).

2 Case Studies

Three typical case studies of rainfall-triggered deep-seated landslides forming natural dams were addressed in this research, namely the large Kuridaira and Akatani landslide dams in Nara prefecture, Japan and a huge landslide dam in Jure village, Sindhupalchok district, Nepal. The large-scale Kuridaira and Akatani landslides created the two largest natural dams among 17 obstruction locations triggered by extreme rainfall during the severe tropical storm Talas in September 2011 (Hayashi et al. 2013). The Jure landslide dam induced by a very high accumulative rainfall in August 2014 is considered the worst rainfall-induced landslide creating the biggest natural dam reservoir in Nepal (Tien 2018). Due to high risks from sequential hazards (like overflows, debris flows, river erosion, dam breach,…), these three landslide dams and their reservoirs were forced under urgent countermeasures by countries’ governments. The details of the three case studies are presented in Table 1.

Table 1 Description of Kuridaira, Akatani and Jure landslide dams

The Kuridaira and Akatani dams were formed by massive movements triggered by heavy rainfall, with a record-breaking cumulative precipitation of about 1800 mm in the period from August 31 to September 4 (Fig. 1). The Kuridaira landslide was approximately 100 meters in depth, 800 meters in length, and 600 meters in width, with a material volume of 23.0 million m³. The Akatani block, which had a landslide volume of 10.2 million m³, was 67 m deep, 1000 m long and 300–500 m wide at its head and toe. The Kuridaira and Akatani sliding blockages created natural reservoirs with heights of 100 m and 85 m and storage capacities of 7.5 and 5.5 million m³, respectively (Tien 2018; Tien et al. 2018). The Kuridaira landslide was characterized by alternating layers of sandstone and shale rocks, as well as clastic deposits. The landslide slid along the northwest dip-slip faults within the shale and sandstone layers or their interbedded layers. While the Akatani slope strata exposed interbedded rock layers of sandstone, sandstone- or mudstone-rich materials formed along the sandstone-rich layers (Fig. 1b, d). The Kuridaira landslide took place at 23:06:13 on September 3 in 100 s. In comparison, the Akatani block failed at 7:22:15 on September 4, lasting for 70 s, with a maximum estimated velocity ranging between 80 km/h and 100 km/h (Yamada et al. 2012; Chigira et al. 2013). The cumulative rainfall amounts triggered Kuridaira and Akatani mass movements were 1516.5 mm and 1746 mm, respectively (Fig. 1e).

Fig. 1

(a, c) Google Earth images and (b, d) slope profiles of Kuridaira and Akatani landslide dams (Sassa et al. 2023), and (e) rainfall data during Typhoon Talas in September 2011 (Tien et al. 2018)

The Jure catastrophic landslide was induced by prolonged rainfall at 2:30 AM local time on August 2, 2014, in Jure Village, Nepal (Fig. 2). The topographical map and slope profile of the Jure landslide are shown in Fig. 2b, c. The sliding is characterized by a rockslide-debris avalanche, moving along bedding-plane faults of weathered metamorphic rocks of phyllite and schist on the slope of an ancient landslide topography (Tien et al. 2021). The landslide disaster claimed 156 people in Ramche, Mankha Tekanpur, and Ghuskun villages. The rapid-moving landslide travelled at a maximum velocity of about 60–70 m/s, swept away and buried communities comprising 60 households in the villages of Iteni and Kagu. Then, the displaced mass rushed downward to Jure village and destroyed 1.0 km of the Araniko National Highway, causing significant disruptions to vital trade activities between Nepal and Tibet, China. The landslide, measuring 1300 m in length and with a volume of 13 million m³, completely blocked the narrow river valley within 2–3 min. The massive sliding created a natural dam that was 700 meters long and 370 meters wide on Sunkoshi River in Jure village. The impounded lake had a maximum depth of 47 m, a length of 3.1 km and a maximum storage capacity of 11.5 million m³ (Fig. 2a).

Fig. 2

(a) Photograph of the Jure landslide dam taken in December 2016 and a google image of the natural lake after damming in August 2014, (b, c) topographic map and longitudinal profile (J-J) of the landslide, (d) rainfall data at the Barhabise station and (e) elevation-volume and elevation-surface area curves of the dam reservoir (Tien et al. 2021)

The impounded water flooded a section of the highway and approximately 100 houses situated along the banks of the Sunkoshi River. Although the Nepal Army took emergency countermeasures against the dam failures and overtopping by excavating the dam body and opening the drainage channels, the reservoir was breached due to the continuous heavy rainfall. The resulting outburst flood not only destroyed the Araniko Highway and Sunkoshi Hydropower Plant but also caused damage to the areas within about 6 km downstream along the river basin (Tien et al. 2021). The landslide was triggered by a prolonged period of heavy rainfall, which accumulated to a high total of 1117 mm from May 29 to August 2 (as shown in Fig. 2d). The relationships between elevation-volume and elevation-area of the Jure dam reservoir are denoted in Fig. 2e.

3 Materials and Method

Ring shear tests were carried out on landslide samples to examine the physical mechanisms responsible for the formation of landslide dams in river valleys. Soil samples were collected from the sliding surface layers of the landslides. These samples were passed through a 2-mm sized sieve and tested using the undrained high-stress, dynamic-loading ring shear apparatus (ICL-2), which has a capacity of 3.0 MPa in normal stress. A graph showing the grain-size distribution of shale sample (K2), sandstone-rich material (A1) and weathering material of schist (J2) of three Kuridaira, Akatani and Jure landslides, respectively, is presented in Fig. 3. For ring shear simulation, the initial stress state of landslide samples due to gravity was calculated from the sliding depth (H), slope angle (θ), and soil unit weight (g), namely normal stress δ₀ = g.H.cos²θ and shear stress τ₀ = g.H.cosθ.sinθ. The initial normal stresses of 1000 kPa, corresponding to 70 m depth of the sliding surface, were employed for the Kuridaira and Akatani landslides. A normal stress of 600 kPa estimated based on the maximum sliding surface depth of 43 m was applied to study the Jure landslide. In this research, pore pressure control tests were performed to indirectly simulate the rainfall-triggered landslides due to the rise of the groundwater table to investigate the physical mechanisms governing the landslide dam formation.

Fig. 3

(a) Grain-size distributions of landslide samples: Kuridaira sample (K2), Akatani sample (A1) and Jure sample (J2)

The LS-RAPID landslide simulation model was employed to simulate the entire process of the formation of three landslide dams, starting from stable slopes, through sliding initiation and motion, and ending with river blockage to create natural lakes behind the dams. The LS-RAPID model integrates the sliding initiation and motion progress from the stable state to failure and eventually reaching a steady state in the post-failure condition (Sassa et al. 2010). This computer model incorporates geotechnical soil parameters measured in the ring shear experiments. Topographical input of this 3D model employed DEM data of three landslides before and after the sliding. In the LS-RAPID model, pore water pressure increment indirectly simulating the rise of groundwater level under rainfall was employed as the landslide triggering factor.

4 Formation Mechanism of Rainfall-Induced Landslide Dams by Ring Shear Tests

4.1 Test Results on Kuridaira and Akatani Landslide Samples

In pore water pressure control tests, the saturated samples of K2 and A1 were first consolidated to normal stress of 1000 kPa and shear stress of 600 kPa and 620 kPa for Kuridaira and Akatani landslides in drained condition. These values correspond to the initial stress states (black line) of the Kuridaira and Akatani landslide samples with an inclination of 31° and 34°, respectively. Then, the simulation of rainfall-induced landslides was reproduced by the pore water pressure increment that indirectly simulated the increasing of groundwater level due to rainfall. The increment rates of pore pressure values were maintained at a constant rate of 1.0 and 0.5 kPa/s for samples K2 and A1, respectively. The results of pore water pressure control tests on samples K2 and A1 are present in Fig. 4.

Fig. 4

Pore-water pressure control tests on samples K2 and A1: (a, c) stress path and (b, d) stresses and pore water pressure in the progress of shear displacement

Test results present that the Kuridaira and Akatani samples failed at the pore pressure increment of about 340 and 370 kPa. These values indicate that the critical pore water pressure ratios (r_u) due to rainfall initiated the Kuridaira and Akatani landslides were 0.34 and 0.37, respectively. The parameter r_u is defined as the ratio between the pore water pressure value at the point of failure and the normal stress. The values of friction angle at the peak are 42.7° for the Kuridaira sample and 42.4° for the Akatani sample, while the corresponding apparent friction angles mobilized at 3.4° and 5.5°. The residual shear strengths are 60 and 94.5 kPa for samples K2 and A1. Both the Kuridaira sample (K2) and Akatani sample (A1) show the high mobility behavior due to the excess generation of pore water pressure, which leads to a significant reduction in shear resistance. As a result, the high-speed landslides were observed during the ring shear simulation. The shear displacement values at the starting and end points of shear strength reduction (D_L and D_U values, respectively) are 3.0 and 1500 mm for sample K2. The D_U and D_L values are 5.0 and 1500 mm for sample A1. The failure of the landslide samples experienced four main periods from stable (pre-failure, D < D_L) to failure (D = D_L), through transient stage (D_L < D < D_U) to steady-state (D > D_U).

4.2 Test Results on Jure Landslide Sample

For the rainfall-triggered landslide simulation, normal stress of 600 kPa and shear stress of 360 kPa representing the initial stress condition of the Jure slope were applied to sample J2 in the drained condition. Next, pore water pressure rose at a rate of 1 kPa/s until the failure of the sample. The failure occurred when the path of the effective stress reached the failure line at the peak friction angle of about 35°. The landslide was triggered by a pore water pressure of 158 kPa which corresponds to a critical pore pressure ratio of r_u = 0.26. It indicates that a high value of pore water pressure resulting from rainfall can cause slope failure (Fig. 5a, b). In the simulation, a rapid-moving landslide was monitored during shearing, which suggests a very high mobility of the sample J2. Since steady-state shear behavior and the value of excess pore pressure were not precisely monitored in this drained condition test. The saturated sample J2 was investigated in an undrained shear displacement control test (Fig. 5c, d). In this test, the sample failure occurred as the path of effective stress reached the failure line at a friction angle at a peak of 34.5°. The shear strength value significantly reduced in the progress of shear displacement due to the excess generation of pore water pressure. A very high pore pressure value was measured at a shear displacement of 20 m. The sample stayed at a low residual strength of 70 kPa and a small value of apparent friction angle of 6.5°. The acceleration of rapid motion at the steady state also indicates the high mobility behavior of the sample. The test result indicates that the values of D_U and D_L are 6 and 6000 mm for sample J2.

Fig. 5

Test results on Jure sample J2: Pore-water pressure control test with stress path (a) and time series data (b); and undrained shear displacement control test with stress path (c), and stresses and pore water pressure in the progress of shear displacement (d)

5 Simulating the Entire Evolution Process of the Three Landslide Dams

The entire evolution process of the three landslide dams is investigated in the LS-RAPID model. The input of soil parameters measured in the ring shear experiments is shown in Table 2. Remarkably, the model employed pore water pressure ratios as a triggering factor of the rainfall-triggered landslide dams. The ratios of 0.34, 0.37 and 0.26 for pore water pressure are set up for the computer simulation of the Kuridaira, Akatani and Jure landslide dams, respectively. The pore water pressure ratios were increased to designed values within 10 s and set constant for the duration of 150 s for all the simulations. The modeling results of the landslide dams induced by rainfall are present in Fig. 6. The result analysis indicates that the critical pore water pressure ratios triggering the Kuridaira, Akatani and Jure sliding were 0.33, 0.36 and 0.22, respectively.

Table 2 Soil parameters in the LS-RAPID model

Fig. 6

Results of computer simulation model of three landslide dams (Tien et al. 2018, 2021)

The entire landslide dam formation process consists of four main periods: stable slope (I), local failures (II), progressive failures at transient period (III), and massive slope movement and river damming (IV). All slopes were stable at the first state if the pore water pressure ratios were lower than critical values. When the ratios of pore water pressure due to groundwater/rainfall reached critical values, local failures occurred in the middle parts of Kuridaira and Akatani slopes and the upper part of the Jure slope. The failures of the Kuridaira, Akatani and Jure slopes extended to neighboring areas, and then gradually expanded to a large area of the slopes. The progressive slope failures were characterized by the downward movement of the upper parts, resulting from the loss of support at the base and the dynamic loading impacts from the upper zones. The landslide accelerated its movement with increasing velocities during the transient stage. After that, the mass movements rushed down the river valleys and collided with the opposite walls. The landslide debris started accumulating on the river floor and then widely spread out in downstream and upstream areas due to the high mobility behavior of the sliding materials. In the final stage, the sliding motions decelerated and soon ceased when the whole mass movement slid down to the valley. The completion of river damming was observed in the narrow V-shaped valleys of the Kuridaira, Akatani and Jure areas. Specifically, the sufficient volume of the massive landslides completely dammed the river to form natural reservoirs in a short period. The results of the computer simulation for the K-K, A-A, and J-J cross-section and run-up point of the three landslides are presented in Fig. 7. The topographic data obtained from the LS-RAPID model that were well compared with the original terrain before the sliding. The run-up points of the three landslides were nearly the same as those drawn from the original DEM data before sliding.

Fig. 7

(a–c) Topographic data of three landslides in comparison with model results and (d) an observed run-up point on the opposite slope in the Jure area (Tien et al. 2018, 2021)

6 Conclusions

This paper presents the investigation of the physical mechanism and entire landslide dam formation process through three typical case studies of Kuridaira, Akatani and Jure slopes. The laboratory tests indicate that the landslides were triggered by an increase in pore water pressure under rainfall conditions. The values for pore water pressure triggering the Kuridaira, Akatani and Jure landslides are 0.34, 0.37 and 0.26, respectively. In the tests, all samples of K2, A1 and J2 show a high mobility behavior because of a significant loss of shear strength from an excess generation of pore water pressure. The mobility behavior was expressed by low mobilized friction angles and residual strength of the landslide samples at the steady state. As a result, the rapid motions were monitored in the rainfall-induced landslide simulation by ring shear apparatus. Notably, the evolution of the three landslide dams was simulated in the LS-RAPID computer model that incorporated geotechnical soil parameters measured from the ring shear tests. The entire evolution of the landslide dam comprises four main stages: stable slope, local failure, progressive failures, massive slope movement and river damming. In the model, three landslides travelled downward to their river valleys at high speeds and created natural lakes. The high mobility behavior of the soil samples characterized by rapid-moving landslides and the sufficient volume of sliding materials played essential factors in the dam formation. The mobile landslides with wide-spreading and rapid motions allow the displaced material blocks to completely block the river channels quickly before their extensive erosion by upstream water flows. The result analysis suggests that landslide mobility is a significant parameter for hazard assessment of landslide dams. In addition to geomorphic features, the high mobility behavior of the Kuridaira, Akatani and Jure landslides was one of the significant contributing factors for the dam formation in the study areas.