Mechanism of a rainfall-induced landslide in a large-scale flume experiment on a weathered granite sand

Introductions A large-scale flume experiment was performed to evaluate the mechanism of landslide occurrence due to rainfall using weathered granite sand. The dimensions of the flume were 9 m (length), 1 m (width), and 1 m (depth). The weathered granite sand from the actual landslide site at Da Nang City, Vietnam was used. The pore water pressure was measured by a pore-water pressure transducer at two depths (middle and bottom) to determine the process of rainwater infiltration into the soil. The surface deformation was measured with extensometers at three positions of the slope. The deformation of the entire slope was determined by the 160 cylindrical-shaped makers evenly spaced in the slope and three cameras. Results The results showed that the rainfall infiltrated into the slope process, increasing from negative pore water pressure to approximately 0. The maximum shear strain contour has been plotted in total and in time increments. The shear band was detected from the time increments maximum shear strain contour. The localization in the shear band formed just before failure. Conclusions To the best of our knowledge, this is the largest scale laboratory test ever conducted to calculate the shear band. Moreover, it was found that the failure occurred when the sand was in an unsaturated phase. Failure does not seem to depend on the increase in pore water pressure but on the maximum shear strain. This feature can be used to explain the phenomenon of landslides that occur even when the groundwater level does not increase but large deformation occurs.


Introduction
Landslides are frequent disasters that happen every year around the world.During the rainy season, landslides often occur in areas that have heavy rainfall.In Central Vietnam, the rainy season from October to December is the time when landslides are common.In the year 2011-2016, a technical cooperation project between Japanese and Vietnamese researchers was conducted to develop landslide risk assessment technology along transport arteries in Vietnam (Tien et al. 2017).There were four main parts in the project including (1) Landslide mapping, (2) Material Testing and Software simulation, (3) Landslide monitoring, and (4) Landslide flume experiments.The project selected a landslide area at Hai Van Mountain in Da Nang City, Vietnam as a pilot study.At the toe of the mountain, the North-South railway passes through Hai Van station.In this study, the landslide is named Hai Van station landslide.At Hai Van station landslides, landslides often occur in the rainy season from September to December (Fig. 1).Small-scale landslides occurred around Hai Van station in 1999Van station in , 2005Van station in , and 2007 (Tien et al. 2015).In the rainy season of 2005, more than 7000 m 3 of rock and soil moved down and covered the railway (Quang et al. 2018).The location of Hai Van railway station is very important.Hence the project chose this landslide point for research.There were some studies on the landslides around this area (Abe et al. 2018;Ha et al. 2018;Quang et al. 2018).These studies focused on topography, geology, hydrogeological structure, soil properties using the ring shear apparatus, and landslide numerical simulation.According to the research results of Abe et al. (2018), the soil in Hai Van Mountain is weathered granite sand.In this study, a large-scale flume experiment was performed to study the mechanism of rainfall-induced landslides using weathered granite sand from the Hai Van station landslide.
Many flume experiments have been conducted both inside the laboratory as well as outside the field in past decades.Some flume experiments were small-scale with a length of 0.50 m (Zhang et al. 2011), a length of 0.60 m (Xu et al. 2022), a length of 1.22 m (Tiwari et al. 2014), a length of 1.50 m (Cogan and Gratchev 2019), a length of 1.80 m (Wang andSassa 2001, 2003).Some flume experiments were large-scale with a length of 9.00 m (Okura et al. 2002;Okada 2014), a length of 20 m (Lee et al. 2021), a length of 21.06 m (Moriwaki et al. 2004).A flume experiment on a natural slope with a length of 30 m was conducted by Ochiai et al. (2004).
The scale of the flume experiment has an influence on the test results.Small-scale experiments have problems with scaling effects and discontinuity effects of sensors and cables (Moriwaki et al. 2004).To maximize geomorphological relevance, landslide experiments must be conducted at the largest feasible scales (Iverson 2015).The flume experiment on a natural slope recreates the natural conditions most accurately.However, monitoring the displacement under the ground is limited in the study on a natural slope.The materials used for the flume experiment were also important.Grain size has a significant impact on the mobility of rainfall-induced landslides (Wang and Sassa 2003).The material taken from the site is the most similar grain size to the actual field material.Okura et al. (2002), Moriwaki et al. (2004), and Okada (2014) used river sand for flume experiments.Lee et al. (2021) used weathered soil for a flume experiment but did not mention the source of material from the actual landslide field or not.In the above flume experiments, there is no research on the mechanism of a landslide using the weathered granite from the actual landslide field.
Furthermore, determining the location of the failure surface is very important in landslides.This failure surface formation is related to the shear band formation in the slope.It is very difficult to observe the shear band formation inside the sliding mass.To estimate the location of the failure surface, several methods have been reported (Jaboyedoff et al. 2020).Some methods have been applied to determine the location of the failure surface, such as a calculation of a shear band direction (Tatsuoka et al. 1990), and obtaining a wetting front (Ahmadi-adli et al. 2017).Besides, many small-size laboratory experiments have been performed to simulate and record the shear band formation.Common methods are direct shear tests (Nitka and Grabowski 2021), triaxial tests (Desrues and Chambon 2002), plane strain tests (Tatsuoka et al. 1990;Alshibli and Sture 2000;Kwak et al. 2020), ring shear tests (Sadrekarimi and Olson 2010), and sandbox experiments (Wolf et al. 2003).These tests applied an external force to produce a deformation in an orientation direction.However, the above studies used small-scale models, and they did not determine the location of the failure surface using image analysis of markers to calculate the maximum shear strain distribution and the direction of zero extension lines.
In this study, a large-scale flume experiment with a length of 9 m, a width of 1 m, and a depth of 1 m was conducted.Artificial rain was sprayed continuously with a constant intensity of 50 mm/h to investigate the failure mechanism of a rainfall-induced landslide.The rainfall intensity is similar to rainfall data at the field.The soil material used for this study is weathered granite which was taken from the landslide area at Hai Van station landslide.Figure 2 shows the flume experiment with the side view and top view before and after the failure.During the experiment, the pore water pressure was recorded using piezometers and a multi-tensiometer.The surface slope deformation was recorded using extensometers.The entire slope deformation was recorded using video cameras and cylindrical-shape makers.
The study analyzed the relationship between the displacement and deformation of the slope with a change in pore water pressure due to rainfall.In addition, the study also shows a method to determine the location of the failure surface by drawing a maximum shear strain contour.

Flume experiment
The illustration and photos of the flume experiment with the monitoring sensors system are shown in Fig. 3.The flume experiment had a length of 9 m, a width of 1 m, and a depth of 1 m.One side of the flume experiment was made of reinforced glass so that the movement of the entire soil mass could be observed from cameras.The 9 m long flume experiment was divided into three sections with different slope angles simulating the natural slope.The top part was 1 m long with a slope of 0 degrees simulating the top of the slope.The middle part was 4 m long with a slope of 34 degrees similar to the slope at Hai Van station landslide.The lower part was 4 m long with a slope of 10 degrees, simulating the gradual shape of the slope.To simulate the phenomenon of rain, a system of 5 nozzles was designed on the roof, 1.6 m from the highest point of the flume experiment.The nozzle system was designed so that the rainfall was relatively uniform along the length of the flume experiment.To regulate the amount of artificial rain, a system of pressure-regulating valves was installed along the water pipe to the nozzles.

Monitoring system
The monitoring system was designed to monitor the change of the pore water pressure, surface deformation, and deformation of the entire slope in the flume as shown in Fig. 3.The length of the flume was divided into sections from A to J with 1 m apart to install the monitoring devices.Table 1 shows the number and measurement parameters of monitoring devices in each section.Twenty piezometers were installed evenly from sections A to J at two depths of 0.30 m and 0.60 m to measure pore water pressure.In addition, a multi-tensiometer was installed at section F at three depths of 0.15 m, 0.30 m, and 0.45 m.
The multi-tensiometer was a newly designed pore water pressure gauge that can measure pore water pressure at multiple depths.The purpose of the multi-tensiometer was to compare pore water pressures measured with two different types of devices versus piezometers.The piezometer is a strain-gauge type that can measure porewater pressure at one depth (18 mm in diameter, 83 mm long, range ± 70 kPa, accuracy = 0.015%).The multi-tensiometer is a piezoresistive silicon pressure type that can measure pore-water pressure at multi-depth (48 mm in diameter, 937 mm long, range ± 100 kPa, accuracy = 1%).
Three extensometers were installed on the soil layer surface at section F, section G, and section H to measure surface deformation.One hundred and sixty markers were installed evenly along the flume with 0.20 m distance horizontally and at depths of 0.10 m, 0.25 m, 0.40 m, and 0.55 m as targets (as shown in Figs. 2, 3).The markers were 0.016 m diameter and 0.06 m length acrylic cylinder and aluminum.The markers were glued with reflective yellow tape so that the video camera could record images during the movement.Three cameras were placed in section C, section D, and section H to record the movement of the soil mass and the markers.The monitoring system was time synchronized between the measuring devices.Data loggers and monitoring data display computers connected to all piezometers, multitensiometers, and extensometers.

Experimental material
The experimental material was weathered granite sand taken from the landslide area at Hai Van station landslide, Da Nang City, Vietnam.The experimental material was excavated on the surface of Hai Van Mountain (Fig. 1).The soil at the site had large particles and even rocks several meters in size.We screened large particles and only used particles less than 20 mm in size for the flume experiment.Figure 4 shows the grain size distribution of Hai Van soil used for the flume experiment after screened.The particle size at 50% passing by mass was 0.429 mm.Mass percentages passing at particle size 0.075 mm was 19.20 (%).The particle size at 10% passing by mass was 0.015 mm.The particle size at 60% passing by mass was 0.622 mm.The coefficient of uniformity was 41.47.

Testing procedure
Before placing soil on the flume, the amount of rain sprayed from the nozzles was adjusted, so that the rainfall was relatively uniform over the entire length of the flume with the required amount of rain.According to the monitoring rainfall data at the Hai Van station landslide (Fig. 5), a rainfall of 50 mm/h was selected for the experiment.
The soil was transported by a crane, spread on the flume in layers, and compacted by using the weight of a person to step evenly on the surface of the soil layers.The thickness of the soil mass covering the slope was 0.60 m.The soil was filled in 4 layers.Each layer was 0.15 m thick.During backfilling, markers were placed in a grid   that was evenly spaced along the slope.These markers moved together with the soil in the sliding mass during the deformation.
Table 2 shows the testing procedure.An initial amount of 50 mm/h rainfall was sprayed for 10 min.Then, the water was allowed to infiltrate the soil for 6 h.The purpose of this initial spray was to give the soil a certain amount of moisture.After 6 h, soil samples were taken for testing at 4 depths in 4 layers at different locations along the flume experiment.Table 3 shows the soil properties of Hai Van sand at the initial condition.The average initial degree of saturation was 27.39%.
At the beginning of the experiment, the rain was sprayed continuously with a constant intensity of 50 mm/h until the end at 8259 s.The change in pore water pressure was recorded by piezometers and multitensiometers.The displacement of the surface was recorded by extensometers.The movement of markers in the moving mass was recorded by video cameras.The last moment that the cameras could record the position of all markers was at 8247 s.The failure occurred from 8247 to 8259 s.
After failure, soil samples were taken for testing at 4 depths in 4 layers at different locations along the flume experiment.Table 4 shows the soil properties of Hai Van sand after the slope failure.After the slope failure, the average degree of saturation was 87.51%.Thus, both before and after slope failure, the soil in the slope was in an unsaturated phase.

Pore water pressure ratio calculation
From the pore water pressure measured by piezometers, the pore water pressure ratio at the depth of piezometers was calculated.The pore water pressure ratio is calculated based on the following equation by Atkinson ( 2007): (1)   where u is the absolute value of pore water pressure at each time.
Δu is the change in the pore water pressure.σ v0 ′ is initial vertical effective stress.γ sat is the total unit weight of soil.γ w is the unit weight of water.h is the depth of piezometers.

Maximum shear strain calculation
From the images recording the movement of markers by the video cameras, the shear strain was calculated.The shear strain calculation would determine the strain localization and the shear band formation during the experiment.Figure 6 shows the Mohr circle of strain, angle of dilation, and zero extension lines (Atkinson 2007) to calculate the shear strain.The strain vector, maximum shear strain, and angles of zero extension lines of the 4-noded quadrilateral element at the position of 4 markers during the failure are calculated based on the following equations: where ε is strain vector, B is strain displacement matrix, d is displacement, ε x is a normal strain in the x-direction, ε y is a normal strain in the y-direction, γ xy /2 is shear strain, α and β are angles of zero extension lines, ψ is an angle of dilation.The value of maximum shear strain γ max /2 calculated by Eq. ( 3) is used to draw contours.Angles of zero extension lines calculated by Eq. ( 4) are used to draw zero extension lines.The length of the zero extension lines is shear strain γ xy /2 at normal strain ε = 0. (

Surface deformation and pore water pressure changing
The surface deformation was measured by extensometers at three sections F, G, H on the slope of 34 degrees.Twenty piezometers (P1-P20) were installed along with the flume experiment to record the pore water pressure inside the slope.Odd number piezometers were installed at a depth of 0.60 m, while even number piezometers were installed at a depth of 0.30 m (as shown in Fig. 3).
Figure 7 shows the surface deformation, the velocity, and the pore water pressure from beginning to end.From the slope surface velocity result (Fig. 7b), the experimental process could be divided into two periods.The first period was the precursory period when the slope moved slowly.That was the period from 0 to 8247 s.The second period was the failure period when the slope moved rapidly.This was the period from 8247 to 8259 s.During the failure period, the highest velocity was at 8254 s.The highest velocity of the extensometers Ex1, Ex2, Ex3 were 0.31 m/s, 0.27 m/s, and 0.23 m/s, respectively.According to Hungr et al. (2014), this landslide velocity is classified as extremely rapid.
In Fig. 7a, from the beginning to 8247 s, the three extensometers began to move gradually from 0 m to approximately 0.20 m.Within three extensometers, extensometer Ex1 at the upper section H had the largest displacement.Extensometer Ex2 at section G and Ex3 at the lower section F, respectively, had a smaller deformation.From 8247 to 8259 s, all 3 extensometers moved rapidly from approximately 0.20-1.20 m at extensometer Ex1, to 1.00 m at extensometer Ex2, to 0.68 m at extensometer Ex3.
The data of 20 piezometers are presented in Fig. 7c, d.The pore water pressure at all piezometers had the same tendency to increase from the negative value to approximately 0 kPa.This increase in pore water pressure represents the process of rainwater infiltration into the soil.Within 20 piezometers, only piezometer P6 did not show a significant change in the measured value during the experiment, probably, due to a malfunction.Therefore, the data of the piezometer P6 had been discarded in Figs.8c, 12c and 13c.In Figs.11 and 14, the data of the piezometer P6 was replaced by the nearest piezometer P4.
In Fig. 7c, d, before failure from 6000 to 8247 s, the pore water pressure at all piezometers was almost unchanged at around 0 kPa.In other words, the effective stress did not increase from 6000 to 8247 s.However, failure occurred at 8247 s.During the failure from 8247 to 8259 s, the pore water pressure at piezometer P10 changed clearly.The pore water pressure changed clearly at piezometer P10 may be due to this point was the intersection of two slopes with different slopes.To clearly see the change of the pore water pressure at each cross-section compared with the surface deformation, the pore water pressure at each cross-section from A to J is plotted in Figs. 8 and 9.At each section from section A to section J, there were two piezometers at two depths of 0.30 m and 0.60 m.During the failure that occurred from 8247 to 8259 s, the pore water pressure at piezometers P1, P2, P3, P4, P5 at sections A, B, C (Fig. 8a-c) and the pore water pressure at piezometers P19 and P20 at section J (Fig. 9j) did not change.The other piezometers from P7 to P18 between section D and section I (Figs.8d, e; 9f-i) showed a slight change from 8247 to 8259 s.The slight change of pore water pressure between section D and section I indicates that the failure area was between section D and section I in the flume.
During the failure period, it is noticed that the pore water pressure at piezometers P8, P10, P12, P14, P16, P18 changes.These are the piezometers placed at a depth of 0.30 m within the failure depth of the slope.The movement of the slope during failure causes the piezometers Fig. 9 Changes in surface deformation and pore water pressure from beginning to end at each section: f at section F, g at section G, h at section H, i at section I, j at section J to also move accordingly, leading to a change in water pressure at the corresponding measurement location.The pore water pressure changes the most at position P10, which intersects the two slopes of 10 degrees and 34 degrees.Soil from the 34 degree slope moves quickly and accumulates at this P10 position, causing the pore water pressure here to change greatly.
To compare the pore water pressure measured by piezometers, a multi-tensiometer was installed at section F. The multi-tensiometer was a newly designed pore water pressure gauge that can measure the pore water pressure at depths of 0.15 m, 0.30 m, and 0.45 m. Figure 10 shows the changes in the pore water pressure and the displacement at section F using a multi-tensiometer, piezometer P11, P12, and extensometer Ex3.Both piezometers and multi-tensiometer devices show the same tendency of the pore water pressure to increase from the bottom of the flume to the surface.The results indicate that the multitensiometer can be a useful device for measuring pore water pressure.
In order to understand the process of changing pore water pressure on the entire flume over time, the pore water pressure contours at different times are plotted in Fig. 11.These contours were drawn based on the value of the pore water pressure measured by piezometers at two depths of 0.60 m and 0.30 m, and the assumption that the pore water pressure at the surface was 0 kPa.From 0 to 5000 s, the pore water pressure in the slope gradually turned from negative to approximate zero.The pore water pressure in the lower part of the 10-degree slope increased first.Then the pore water pressure in the higher part of the 34-degree slope increased.From 6000 s to before failure at 8247 s, the pore water pressure was almost the same.During the failure period from 8247 to 8259 s, the pore water pressure was also almost the same.It seems that the pore water pressure before the failure from 6000 to 8247 s did not affect the failure from 8247 to 8259 s (as shown in Figs.7c, d and 11).
From the pore water pressure at each depth of piezometers, the pore water pressure ratio was calculated using formula (1).Figures 12 and 13 show the changes in the pore water pressure ratio at each section from beginning to finish.Only the pore water pressure ratio at piezometer P17 increased from the negative value to approximately 1. Others the pore water pressure ratio increased from the negative value to approximately 0. Figure 14 shows the pore water pressure ratio contours at different times.Over the entire slope, from 6000 s to before failure, the pore water pressure ratio was approximately zero and did not increase.However, slope failure occurred at 8247 s.It means without the decreasing effective stress, slope failure occurred.In other words, the effective stress did not cause slope failure.Another study by Cuomo et al. ( 2021) also shows that the increase in pore water pressure is not the cause of slope failure.Both results of Cuomo et al. (2021) and this study are different from some previous studies that have shown failure to occur when the pore water pressure increases (Iverson et al. 2000;Moriwaki et al. 2004;Okura et al. 2002).

Maximum shear strain changing
At 8247 s, some markers started to move quickly and could not capture the location from the video camera (Fig. 15 and Table 2).Therefore, the shear strain and maximum shear strain calculation based on the position of the markers were also calculated only up to the last time at 8247 s. Figure 16 shows the deformation at all markers on the slope recorded by video cameras from the beginning to before failure.Before the failure occurred, the soil at the 34-degree slope between section E and section I had moved significantly.The soil at the 10-degree slope between the A and E sections as well as the top soil between the I and J sections mainly subsides.At the cross-sections C, E, G, and I where there are steel bars and no markers, the displacement is taken to be the same as that of the nearest markers to the left.
From the movement of markers over time, formulas (2), (3), and (4) were used to calculate the maximum shear strain and zero extension lines.The maximum shear strain was calculated at the 4-node position of an element consisting of 4 markers.The maximum shear strain value at a node was calculated as the average value of the maximum shear strain at that node on the surrounding elements.
Markers were placed from a depth of 0.10 m to a depth of 0.55 m (Fig. 3).Because there is no marker on the slope surface, the maximum shear strain calculation nodes were calculated from a depth of 0.10 m and deeper.Figure 17 shows the changes in the maximum shear strain at each section from section A to section J at each depth of the marker's position.The maximum shear strain values depend on the depth at each section.The maximum shear strain at shallow depth increased earlier and higher than the maximum shear strain at deeper depth.Maximum shear strain at a depth of 0.10 m had a different variation compared to other depths.At other depths, maximum shear strain increased or remained almost unchanged.But at a depth of 0.10 m, there are periods when the maximum shear strain decreases over time.The decrease in maximum shear strain can be explained by the lack of markers at the surface to a depth of 0.10 m.In addition, the value of the maximum shear strain at each node was calculated as the average value of the maximum shear strain at that node on the surrounding elements.Therefore, the maximum shear strain value at a depth of 0.10 m is unusually different from that at other depths.In order to understand the process of the maximum shear strain's change on the entire flume over time, the maximum shear strain contours were drawn.These contours were drawn based on the value of the maximum shear strain at the depths of 0.10 m, 0.25 m, 0.40 m, and 0.55 m of markers and the assumption that the maximum shear strain at the bottom 0.60 m was 0. Two cases of calculating and drawing the maximum shear strain contours were analyzed.
The first case was total maximum shear strain contours from the start to the points in time before the failure.Figure 18 shows the progressive failure.From the start to 8247 s before failure, the maximum shear strain increased gradually from 0 to approximately 0.5.The highest maximum shear strain appeared the earliest between section H and section I where the failure surface originated.The maximum shear strain then increased at other locations on the body of the flume.At 8247 s before the failure occurred, the highest maximum shear strain was concentrated between section H and section I where the failure surface was generated.
The second case was the maximum shear strain contours at every time increment.Figure 19 shows the time increment of maximum shear strain contours at every step.From the start to 8000 s, the maximum shear strain increased mainly in the 34-degree slope.At different time intervals, the maximum shear strain increased at different positions on the 34-degree slope.The maximum shear strain from 8000 to 8247 s shows that the highest maximum shear strain areas were concentrated at a continuous long and narrow area between section G and section I.This highest maximum shear strain is in a continuous long and narrow area called the shear band.The position of the shear band between section G and section I, and the position of the failure surface detected from the video camera is compared in the general discussion.
Comparing Figs. 11,14,18 and 19,it shows that landslides can form in the following two stages corresponding to changes in pore water pressure and slope movement.The precursory period of rain causes pore water pressure to increase at the entire slope (from 0 to 5000 s).After that, the rain continued (from 6000 to 8000 s) without increasing pore water pressure, but the movement became more obvious, starting from section I and gradually developing to section F. The next stage is just before failure occurred (from 8000 to 8247 s), pore water pressure did not increase, but deformation increased and shear band formed between Fig. 12 Changes in pore water pressure ratio from beginning to end at each section: a at section A, b at section B, c at section C, d at section D, e at section E, f at section F, g at section G, h at section H, i at section I, j at section J sections I and F. From the results of this study, it shows that the mechanism of landslides caused by rain is an increase in pore water pressure leading to increased slope deformation.When the slope deformation increases, a shear band forms, then landslides occur.
To compare the position of the failure surface formation with the position of the shear band and zero extension lines, a further analysis was conducted.According to Atkinson (2007), shear zones usually appear to have no thickness so they are called failure planes or failure Fig. 13 Changes in pore water pressure ratio from 8000 s to end at each section: a at section A, b at section B, c at section C, d at section D, e at section E, f at section F, g at section G, h at section H, i at section I, j at section J surfaces.Since the length of the failure surface remains constant because the material on either side of the failure surface is rigid, it is a zero extension line and its direction is given by Eq. ( 4).By recording the slope deformation by the camera, we were able to determine when the failure occurred and the position of the failure surface at the time of highest velocity.At the highest velocity, it was possible to clearly define the boundary between the deformation and non-deformation parts.That was the position of the failure surface.The result is presented in Fig. 20.It shows the failure surface position compared to the position of the shear band and zero extension lines at 8247 s before the failure occurred between section G and section I. Figure 20a shows the photo of the flume between section G and section I at the highest velocity at 8254 s.The blue dash line is the failure surface detected by the video camera.Figure 20b shows the increment of maximum shear strain contour between 8000 and 8247 s before the failure in comparison with a determination of the failure surface at 8254 s.The determination of the failure surface from the photo shows that the failure surface was along the highest maximum shear strain area before the failure at 8247 s. Figure 20c shows the zero extension lines and the shear strain at 8247 s in comparison with a determination of the failure surface at 8254 s.Before failure occurred, the zero extension lines with the highest shear strain were connected.The determination failure surface from the photo shows that the failure surface was along the line that connected zero extension lines with the highest shear strain.This is the largest scale laboratory test ever conducted to calculate the shear band.Previous studies were only able to use small experimental equipment such as direct shear test (Nitka and Grabowski 2021), triaxial tests (Desrues and Chambon This flume experiment monitored both surface deformation, deformation in the slide mass, and pore water pressure.The experiment observed the entire deformation process in precursory period and failure period continuously.This kind of monitoring is impossible to perform on actual slopes in the field when the entire slope deformation in the ground cannot be observed.
Experimental results show that the formation of deformation is due to increased pore water pressure when it rains.However, when failure occurs, it is not due to increased pore water pressure but due to increased deformation.This experimental result is valuable in explaining landslides phenomena that occur when the groundwater level does not increase.The reason is that the increased large deformation forms a shear band and failure occurs.

Conclusions
This is the first study on the large-scale flume experiment using weathered granite sand from the actual landslide site.The pore water pressure was measured by a pore-water pressure transducer at two depths (middle and bottom) to determine the process of rainwater infiltration into the soil.The surface deformation was measured with extensometers at three positions of the slope.The displacement of the entire slope was Fig. 16 Deformation at all markers on the slope recorded by video cameras from beginning to before failure Fig. 17 Changes in maximum shear strain at each section: a at section A, b at section B, c at section C, d at section D, e at section E, f at section F, g at section G, h at section H, i at section I, j at section J determined by the 160 cylindrical-shape makers evenly spaced in the slope and three cameras.The conclusions of this study are as follows: 1.In the flume experiment, the extensometer was used to determine the slope surface deformation.From the deformation data, the slope surface velocity was calculated.From slope surface velocity results, the experimental process could be divided into two periods.The first period was the precursory period when the slope moved slowly.The second period was the failure period when the slope moved rapidly.2. The piezometer results showed that the rainfall infiltrated into the slope process.The pore water pressure at all piezometers had the same tendency to increase from the negative value to approximately 0 kPa.During the precursory stage before the slide, the pore water pressure increased to near zero.However, effective stress not increasing or not changing, but failure occurs.This result differs from other studies that have shown failure to occur when the pore water pressure increases.3. The movement of the makers throughout the sliding mass was used to calculate the shear strain.The maximum shear strain contour has been plotted in time increments.Before the failure occurred, a shear band was formed and detected from the maximum shear strain contour.The position of the shear band coincides with the position of the failure surface.We successfully plotted the maximum shear strain field in a large-scale flume experiment.To the best of our knowledge, our study is the largest scale experiment ever conducted to calculate the shear band formation from the maximum shear strain field.This experi-Fig.18 Total maximum shear strain contours from beginning to before failure in the entire slope mental result is valuable in explaining landslides phenomena that occur when the groundwater level does not increase.The reason is that the increased large deformation forms a shear band and failure occurs.4. By recording the slope deformation by the camera and extensometer, we were able to determine when the failure occurred and the position of the failure surface.From the deformation data recorded by the extensometer, the velocity was calculated.At the highest velocity, compared with photos recorded by the camera, it was possible to clearly define the boundary between the deformation and non-deformation parts.
That was the position of the failure surface.5.The position of the failure surface, maximum shear strain, and zero extension lines have coincided.
Therefore, the calculation of the maximum shear strain contour could be a method to determine the failure surface.6.In this flume experiment, we used materials from the field and removed large particles, using rainfall intensity similar to rainfall data at the field.This experiment can be considered to be similar to the actual field.Besides, we used a high-accuracy and timesynchronous measuring device that can determine the displacement and change of pore water pressure continuously during the whole experiment.Therefore, the flume experiment results of good quality can be considered for the actual mechanism of landslide occurrence in the field.

Fig. 1
Fig. 1 Hai Van station landslide in Da Nang city, Vietnam

Fig. 2
Fig. 2 Photos of the flume experiment.a Side view at the beginning.b Top view before failure.c Top view after failure

Fig. 3
Fig. 3 Location and photos of the sensors at the flume experiment

Fig. 4
Fig. 4 Grain size distribution of Hai Van soil used for flume experiment

Fig. 6
Fig.6Mohr circle of strain, angle of dilation, and zero extension lines(Atkinson 2007), where ε x is a normal strain in the x-direction, ε y is a normal strain in the y-direction, γ xy /2 is shear strain, α and β are angles of zero extension lines, ψ is an angle of dilation

Fig. 7
Fig. 7 Changes in: a surface deformation, b velocity, c, d pore water pressure from beginning to end

Fig. 10
Fig. 10 Changes in pore water pressure and surface deformation at section F: a from beginning to before failure, b enlarged portion from 8000 to 9000 s

Fig. 11
Fig. 11 Pore water pressure contours at different times in the entire slope

Fig. 14
Fig. 14 Pore water pressure ratio contours at different times in the entire slope

Fig. 19
Fig. 19 Increment of maximum shear strain contours in the entire slope

Table 1
List of monitoring devices in sections A to J

Table 2
The timeline of the flume experiment

Table 3
Soil properties at initial condition

Table 4
Soil properties after slope failure