Keywords

1 Introduction

Landslides are one of the pervasive natural hazards occurring on the planet. Each year these landslides cause not only billions of cash in economic losses but also a priceless socio-impact. In particular, Sri Lanka ranks the fourth place and the seventh place according to landslide events per 1000 km2 and fatalities per million inhabitants (Gómez et al. 2023). In recent years, numerous measures, particularly the implementation of landslide early warning systems, have been taken to manage and minimize the damage brought on by relentless landslides. These developments are vital because early predictions are not only a viable option but also crucial for ensuring the safety of inhabitants, as it enables proactive evacuation plans and safeguard the lives at risk.

Certainly, rainfall is one of the most important external factors that lower soil suction and shear strength, which in turn leads to shallow landslides. However, the absence of a more thorough hydrological process involving groundwater flows means that temporal predictability still poses a challenge. Understanding the complex behavior of pre-existing groundwater flow and their recharge/surcharge during a rainfall event is essential for strengthening the landslide early warning mechanism. Hence, this study aims (1) to validate the functioning of the newly developed centrifuge chamber to reproduce both groundwater flow and rainfall infiltration and (2) to examine the initial condition of the soil-bedrock interface, influenced by the existing groundwater flow, on the slope's response to subsequent rainfall infiltration by centrifuge modeling.

2 Experimental Setup

2.1 Centrifuge Model Testing

Many centrifuge model studies have focused on slope behavior under rainfall infiltration, except a few recent studies incorporated groundwater flow effects in the simulation (Askarinejad et al. 2012; Lucas et al. 2020; Take et al. 2015). Therefore, there is an essential need to comprehensively discuss the hydromechanical behavior of slopes subject to pre-existing groundwater flow and rainfall infiltration.

A novel soil container was designed to conduct unsaturated slope stability analysis using centrifuge modeling. This container was capable of simulating both rainfall and groundwater flow conditions independently, allowing for parallel control of these two crucial occurrences. The schematic diagram of the centrifuge container, with dimensions of 800 × 150 × 600 (mm), is illustrated in Fig. 1. Scaling factors specific to this study can be found in Table 1 in which “N” represents the enhanced gravitational acceleration. It is important to note that all experimental results presented in this research are scaled models unless explicitly mentioned otherwise. The model testing was conducted at the geotechnical centrifuge facility at the Disaster Prevention Research Institute (DPRI), Kyoto University, under 50 g conditions. To simulate groundwater flow conditions, the water in the drainage tank was pumped up to the intermediate water tank via an overhead water storage tank using a container-mounted pump. To initiate the groundwater simulation, electrode sensors were activated, and water was subsequently released through the drainage slits located along the right-side boundary wall.

Fig. 1
A schematic of a centrifuge container. It consists of an air atomizing nozzle system for rainfall simulation, a porewater pressure transducer, an intermediate water tank to control the groundwater table, drainage slits, groundwater simulation, drainage hole, and drainage tank.

Schematic diagram of the centrifuge container

Full size image
Table 1 Centrifuge scaling laws applied in the experiments
Full size table

2.2 Slope Construction

To analyze the development of transient porewater pressure and deformation before failure, the slope configuration was carefully established, as shown in Fig. 1. The soil slope’s dry density was maintained at 1.5 g/cm3. Air-dried soil with a gravimetric water content of 10% was mixed and homogenized for 24 hours before slope construction. The wet tamping method was employed to construct the slope layer by layer, with each layer being 20 mm thick.

During the construction process, porewater pressure transducers (PPTs) were strategically embedded at specific locations, as depicted in Fig. 1. However, it is important to note that the available PPTs were only capable of measuring changes in positive porewater pressure. Additionally, markers were positioned in such a way that facilitated the examination of the slope's deformation behavior throughout the progression of the landslide.

2.3 Slope Material

In this study, the experiments were conducted using a silty sand soil commonly known as “Masado soil” readily available in Japan. The index properties of the soil, along with the parameters of the soil water characteristic curve (SWCC), were determined. The SWCC parameters were established based on the van Genuchten model (van Genuchten 1980). The material properties of the soil are presented in Table 2.

Table 2 Material properties of Masado soil
Full size table

2.4 Groundwater Flow Simulation

As mentioned earlier, one of the main objectives of this study is to validate the functionality of the newly developed centrifuge soil container. To achieve this, five identical unsaturated soil slopes were tested. Before inducing failure in the slopes through the combined effects of groundwater flow and rainfall, it is required to establish a steady-state groundwater flow condition. To accomplish this, the water level in the intermediate tank was controlled until porewater pressure transducers placed on the soil–bedrock interface produced steady values. The experimental procedure followed in this study to set up steady-state groundwater flow is illustrated in Fig. 2. Following similar steps, it was expected to achieve similar porewater pressure (PWP) values across corresponding PPTs in all five test cases.

Fig. 2
A flowchart of the steps in the study. Water stored in the drainage tank moves to an overhead water storage tank. The water then moves into an intermediate water tank, released to slop via drainage slits, porewater pressure values are observed, and rainfall is applied to values.

Experimental steps followed in this study to set up steady-state groundwater flow

Full size image

Firstly, the water stored in the bottom drainage tank was pumped to the overhead water tank using the pump mounted on the chamber. The overhead water tank has only a capacity of 0.4 l. Therefore, an overflow line was connected back to the bottom drainage tank. The water stored in the overhead water tank can be discharged to the intermediate water tank by activating the electrode sensors and the water level can be maintained accordingly. Once the intermediate water tank is filled, water will be conveyed to the unsaturated soil slope through drainage slits. The water level in the intermediate tank is getting reduced during this process and once it reaches below the opening of the drainage slit, electrode sensors will be automatically activated to refill. This repetitive process facilitated obtaining steady values in PPTs and attaining steady–state groundwater conditions.

The water level was adjusted in the intermediate water tank as described above to generate a steady–state groundwater flow. However, it was understood that maintaining a constant water level (constant PWP) in the intermediate water tank was not a successful approach to establishing a steady-state groundwater flow. If the water level was maintained at a constant low level, it was hard to obtain steady–state conditions whereas, if the water level was maintained at a higher level, the pressure exerted on the right side of the slope was greater. Eventually, it resulted in undesirable movements in the slope. Therefore, as illustrated in Fig. 2, the water level in the intermediate water tank had to adjust to obtain steady and similar values in the PPTs placed along the soil-bedrock interface. It should be noted that this observation-based method, although utilized, cannot be considered a standard approach for achieving steady-state groundwater flow.

3 Validating the Functioning of the Centrifuge Chamber

3.1 Porewater Pressure Variation during Seepage

In Fig. 3, the PWP in the intermediate water tank varied mostly between 80 and 90 kPa during each test case. However, the time taken for the groundwater flow to reach a steady—state showed significant variations across the cases. This phenomenon can be attributed to two factors. Firstly, the system was not set up to change the water level spontaneously in the same manner for each case. Additionally, although the attempt was to prepare identical slopes using the wet tamping method, it is possible to have non-uniform conditions in each case. Consequently, the response of the PPTs could not be identical, and different test cases required varying time spans to exhibit similar PWP responses. Figure 3 additionally illustrates the time histories of PWP response for PPT 2 in five test cases. It shows that the variation of time needed to reach steady–state conditions vary significantly. However, five cases show nearly identical PWP values obtained at the steady—state. Furthermore, it was understood that a more gentle variation of water level in the intermediate tank is preferred to generate a saturated soil layer on top of impermeable bedrock rather than changing the water level frequently. This procedure permits enough time for the slope to get saturated and thereby creates a more uniform groundwater flow.

Fig. 3
A multi-line graph of porewater pressure in the intermediate water tank and porewater pressure variation versus time. The plots of tests 1 to 5 begin at (0, 70) and follow overall increasing trends. The plots of P P T 2 1 to 5 begin at (0, 0) and follow overall increasing trends. Values are estimated.

Porewater pressure response while establishing a steady state groundwater flow

Full size image

3.2 Porewater Pressure Variation During Rainfall

After PPTs showed enduring values for approximately 100 s, rainfall was applied on the partially saturated slopes by maintaining the groundwater flow. Figure 4 depicts the PWP distribution of PPT 2, 3, and 4 during rainfall infiltration until the failure initiated in each case. -100 s to 0 s represents the time duration of steady–state groundwater flow before the rainfall. Notably, PPT 3 consistently showed higher values compared to PPT 2 and 4, which was observed across all cases. This discrepancy can be attributed to the boundary conditions adopted in these experiments. The coefficient of variation (standard deviation/ average) (CV) was utilized to examine the uniform response of the system of each PPTs as illustrated in Fig. 5. It shows that PPT 2 and 3 have CV values around 0.1 during both steady–state conditions and transient seepage conditions which suggest a high degree of uniformity that aligns with PWP variation. However, a thorough assessment of the PWP variation of PPT 4 in each test shows high variability and this observation was confirmed with relatively higher CV values with a moderate level of uniformity. Even though the variation lies inside the acceptable range, it is necessary to maintain cautious experimental steps to minimize inconsistency.

Fig. 4
3 line graphs of pore water pressure versus time plotted for 5 values of P P T 2, P P T 3, and P T 4. All the plots in the graphs follow overall upward trends.

PWP distribution of PPT 2, 3, and 4 during steady-state groundwater flow and rainfall infiltration until the failure

Full size image
Fig. 5
A multi-line graph of the coefficient of variation versus time. The plots of P P T 2, P P T 3, and P P T 4 follow overall decreasing trends with multiple sharp peaks and dips in between.

Coefficient of variation of PWP distribution of PPT 2, 3, and 4

Full size image

At 0 s, rainfall was applied on the slope. Following the introduction of rain, all the PPTs shown in Fig. 4 exhibited simultaneous responses at around 40 s. However, in the presence of pre-existing groundwater flow, the wetting front and saturated flow quickly propagated throughout the slope, resulting in a rapid response to the rainfall. All the slopes started showing deformation while increasing PWP and eventually failed around 100 s.

3.3 Evolution of Failure

Figure 6 compares the horizontal displacement of marker no 06 (M6) in each test, and in general, the markers’ displacement shows a similar deformation behavior. Therefore, it implied that five identical cases exhibited comparable landslide characteristics. The failure evolution of each case is presented in Fig. 7, which compares the slopes’ initial condition at −100 s and the failure initiation around 100 s. At −100 s, there were some initial movements of about 4 mm in all cases which were initiated as a result of groundwater flow conditions. However, once the groundwater flow became steady, there were no signs of propagating the movements and the slopes were reinstated their stability. This observation can be further validated using the variation of CV value, which shows a constant value with high uniformity as presented in Fig. 8. With the increase of PWP, the failure surface developed during the tests showed that instability propagated from the steep slope to the mid-slope area. The uniformity during deformation is in acceptable range even during landslide progression as shown in Figs. 7 and 8.

Fig. 6
10 photographs of 5 test cases minus 100 seconds and at 100 seconds. The location of M 6 is marked in all 5 test cases at t equals to minus 100 seconds.

Evolution of failure in each test case; location of marker no 06 (M6) is shown

Full size image
Fig. 7
A multi-line graph of horizontal displacement versus time. The plots of 5 tests of M 6 begin close to (negative 100, 4), move parallel to the X axis till (75, 4), and finally, move with positive trends. Values are approximated.

Horizontal displacement of M6 in each test case

Full size image
Fig. 8
A line graph of the coefficient of variation versus time. The plot of M 6 begins at (negative 100, 0.1), moves parallel to the X axis till (25, 0.1), increases to (50, 0.15), dips to (75, 0.05), and peaks at (125, 0.2). The values are approximated.

Coefficient of variation of horizontal displacement of M6

Full size image

The landslide progression observed can be characterized as a translational type of failure compared to a rotational type of failure, mostly occurring as a thick slice of failure in the steep slope as a result of rainfall infiltration. The unstable mass resulting from the tests was voluminous; therefore, the damage that happened under these conditions would be massive.

4 Influence of the Initial Groundwater Flow on Landslide Initiation

Once the functions of the new device were confirmed another two tests were undertaken to assess the slopes’ behavior with (Test 06) and without (Test 07) pre-existing groundwater flow. The objective is to examine the initial condition of the soil-bedrock interface, influenced by the existing groundwater flow, on the slope's response to subsequent rainfall infiltration by centrifuge modeling.

Test 06 was compared with Test 07 to critically discuss the impact of pre-existing groundwater flow on landslide initiation. The unsaturated soil slope in Test 06 was first subjected to pre-existing groundwater flow before being exposed to rainfall infiltration following the method discussed previously. Once the slope reached a steady–state groundwater flow, the PPTs along the soil–bedrock interface (PPT 02 to PPT 06) showed constant values. This has separated the initial unsaturated slope into a saturated layer and an unsaturated layer. In Test 07, the slope stayed completely unsaturated before applying rainfall. Therefore, these two tests allowed for assessing the significance of pre-existing groundwater flow on landslide initiation.

4.1 Porewater Pressure Variation

The variations in the PWP are depicted in Fig. 9. The dashed lines represent the PWP variation in Test 06 and the continuous lines represent the PWP variation in Test 07. Additionally, Fig. 9 shows the variation of PWP in PPT 08 which was placed inside the intermediate water tank of Test 6.

Fig. 9
A multi-line graph of porewater pressure versus time. The plots of P T 02 to 08 of 6 and P P T 02 to 06 of 7 follow overall increasing trends from the onset of rainfall.

Porewater pressure evolution in Test 06 and Test 07

Full size image

One of the major observations was that PPTs in Test 06 responded simultaneously to rainfall (red circle) whereas in Test 07 it clearly showed a sequential response (black circle) (Jayakody et al. 2023a). The possible reasons behind the simultaneous response to rainfall would be the presence of a capillary zone eased the advancement of the wetting front from top-to-bottom saturation over the saturated layer and thereby, the possibility of creating a preferential flow quickly. Further, this observation is a substantiation of uniform infiltration of rainfall to the unsaturated slope. The simultaneous porewater pressure response throughout the slope and quick reaction have made the slope fail much earlier in Test 06 compared to the slope in Test 07. However, in Test 06, the sequential response of PPTs was observed. This is the typical response to rainfall on initially unsaturated slopes, with the advancement of the wetting front, followed by saturated transmission of water flow, and the development of porewater pressure starting from the toe area of the slope. This comparison has highlighted how the change in initial conditions by introducing groundwater flow on soil—bedrock interface could change the response of PWP with rainfall infiltration.

As shown in Fig. 10, PPTs began to react with a rate (∆PWP/∆t) marginally above 0.2 when rainfall infiltration in Test 06 reached the groundwater surface, which was only after 10–20 s. However, later the rate further increased up to 0.35 around 60 s to 80 s which suggests a form of preferential flow paths from the slope surface to the groundwater flow. Afterward, the rate mostly stayed around 0.3 until the slope reaches failure.

Fig. 10
A multi-line graph of the rate of porewater pressure change versus time. The plots of P P T 03 6, P P T 04 6, PP T 05 6, P P T 03 7, P P T 04 7, and P P T 05 7 begin at (negative 120, 0) and move with overall positive trends.

Rate of porewater pressure change in Test 06 and Test 07

Full size image

In contrast to Test 6, PPT 3 took around 50 s to respond to rainfall infiltration at first followed by PPT 4 and 5. At that instance, the rate showed a rate high as 0.4 but it did not continue to be stable with time and shows a rapid declining trend. Except at the time of generating a PWP on soil—bedrock interface, the rate was maintained in between 0.2 and 0.3. This observation was not only in PPT 3 but also in PPT 4 and 5 which eventually shows a less rate compared to the rate in Test 6. Therefore, if an unsaturated slope with a groundwater flow is exposed to rainfall infiltration, there is a high tendency for the rapid development of PWP, which could accelerate the initiation of a landslide. This conclusion can be drawn by looking at the PWP variation and rate of PWP change in Tests 06 and 07.

4.2 Evolution of Failure

The displacements of the markers recorded by the high-speed camera during centrifuge experiments were used to assess the evolution of failure and further quantitative measures. During the failure initiation and follow-up failures, two failure surfaces had very distinct features. In contrast to Test 07's initial failure surface, which resembles a rotational sliding failure, Test 06's initial failure surface is a more extensive translational type sliding failure surface, as shown in Fig. 11. A similar pattern was seen in the progressive failures that occurred after the initial failure, with Test 06 showing considerably greater progressive failures than Test 07. Further, vectors of total displacements at failure initiation in both test cases are also portrayed in Fig. 11 which shows the developed failure surfaces. In Test 06, a dense displacement field was formed not only in the steep slope area but also towards the mid-slope area around 110 s. However, Test 07, clearly shows that a thick localized failure surface has evolved in the top part of the steep slope around 180 s. This concentration of vector arrows is a contrasting behavior with the failure surface evolved in Test 06. With Test 06 revealing an initial failure surface of notably faster, greater dimensions and volume, the evolution of failure as described by physical modeling became fascinating. Even though debris flow and landslide runout were not covered in this article, Test 06 had a lot more failed volume than Test 07. Therefore, it is critical to have a comprehensive understanding of how groundwater flow and rainfall interact to analyze not just the onset of landslides but also their runout.

Fig. 11
4 scanned structures and 2 Y versus X line graphs of M 5 to M 12. The plots in the structures and graphs follow positive trends.

Evolution of failure

Full size image

4.3 Landslide Mobility

As shown in Fig. 12, the displacement during landslide progression was compared in order to examine the mobility of landslides. The comparison of the horizontal displacement between Tests 06 and 07 demonstrates that the markers on both occasions had comparable displacements at a relatively low rate of up to 60 s. However, markers in Test 06 then started to move rapidly and exhibited longer travel distances with time. The simultaneous development in the saturation profile in Test 06 caused markers 5, 6, 11, and 12 to exhibit synchronous movements. However, in Test 07 in particular, markers 05 and 06 demonstrated movement while markers 11 and 12 showed slight movement at the time of failure initiation.

Fig. 12
2 multi-line graphs of horizontal displacement versus time. The plots of tests 06 and 07 of M 5 M 6 and M 11 and M 12 follow increasing trends. The rate of displacement of test 06 is greater than th displacement of test 07.

Horizontal displacement of markers

Full size image

Following the method described in Jayakody et al. 2023a, b, deviatoric strain analysis was carried out to further investigate the landslide mobility of Test 06 and Test 07. The deviatoric strain profiles were calculated employing horizontal and vertical displacements captured. Figure 13a depicts the deviatoric strain profile for Test 06 and it clearly presents that instability was spread over the steep slope as well as the mid-slope. Figure 13b illustrates the deviatoric strain profile for Test 07 and it is apparent that the initial failure surface is a thick slice localized towards the steep slope. The evolution of failure was assessed using not only physical observations but also quantitative measurements using displacement vectors, markers’ displacement distribution, and deviatoric strain profiles. All this evidence guided a very conclusive decision that the impact of pre-existing groundwater flow could create a massive landslide compared to a landslide that occurred only by rainfall infiltration. Therefore, landslide mobility is much higher when the slope has an integrated effect of pre-existing groundwater flow and rainfall infiltration compared to when the slope is only exposed to rainfall infiltration.

Fig. 13
2 contour plots of Y versus X of tests 06 and 07. The plots begin around (100, 0) and increase till (470, 175). Values are approximated.

Deviatoric strain graphs of (a) Test 06 and (b)Test 07

Full size image

5 Conclusions

A newly developed centrifuge model container was used to investigate the importance of incorporating pre-existing groundwater flow in landslide initiation and its characteristics. The study aimed to validate the model container which can reproduce groundwater flow and rainfall as independent functions. Five tests were conducted under identical conditions to evaluate and analyze landslide behavior by monitoring PWP development and failure evolution. The model container demonstrated the capability to replicate comparable results under similar conditions consistently both in PWP generation and subsequent failure mechanism.

The influence of the integrated effect of groundwater flow and rainfall was emphasized by comparing Test 06 and Test 07. Analysis of PWP distribution graphs revealed that the simultaneous response of PPTs during rainfall infiltration resulted in voluminous landslides compared to those triggered solely by rainfall. Further, it was noticed that under the integrated conditions landslide initiation was accelerated. These findings highlighted the necessity of incorporating groundwater flow effects on landslide early warning systems especially to improve the temporal predictability.