Stability Monitoring of Embankments Constructed by Volcanic Coarse-Grained Soil in Snowy-Cold Regions

Geodisasters generated the most serious damage in volcanic ground and slopes in Hokkaido, Japan (Kawamura et al. 2019). It is known that the failures were caused by the specific performance and topography of volcanic soils defined as problematic soils. In addition, it has been reported that many cases of slope failure and landslide that were caused by rainfall and snowmelt occurred mainly from spring to summer, for example, the slope failure of a cut slope on the Hokkaido Expressway in 1999 (Kuromatsunai town, Hokkaido, Japan) and the landslide of a natural slope on the Kiritachi Mountain Pass on Route 239 in 2012 (Tomamae town, Hokkaido, Japan). In particular, the failure of a cut slope on the Hokkaido Expressway in 1999 was caused by rainfall and snowmelt after freezing and thawing. In this study, the authors focus on the rainfall-induced failure of embankments subjected to freezing and thawing among several patterns of slope failure.

A significant amount of research has been reported on the slope instability and the failure mechanism due to rainfall. Rahimi et al. (2011) clarified the effect of antecedent rainfall patterns on rainfall-induced slope failure. Kim and Lee (2010) monitored rainfall infiltration characteristics of a compacted roadside slope and evaluated this using examples from several one-dimensional infiltration models.

In addition, recent studies have provided some results on volcanic soils and slopes that have the particular characteristics of their mechanical behavior. Cascini et al. (2010) proposed a geotechnical model for the failure and postfailure stages of landslides of the flow type in the Campania region, Italy. Comegna et al. (2016) carried out a field hydrological monitoring of pyroclastic deposits for mountainous areas in the Campania region and investigated the seasonal variation in soil suction and water content.

However, although a number of geotechnical studies on issues caused by freezing and thawing and frost heaving have been conducted (e.g., Xu et al. 2011; Zhang and Hulsey 2015), in particular, analytical and in situ experimental approaches that investigated the instability of slopes by freezing and thawing are limited (e.g., Harries and Lewkowicz 2000; Yu et al. 2016; Zwissler et al. 2016; Flynn et al. 2016). Among this research, for example, Xu et al. (2011) investigated the effects of frozen soils on seismic site responses in Anchorage, Alaska. Zwissler et al. (2016) presented that the main control mechanism for slope recession in Calvert Cliffs was freeze–thaw events, and presented a method to quantify freeze–thaw effects.

The authors have investigated the failure of volcanic slopes induced by rainfall and its mechanisms in previous studies (e.g., Kawamura and Miura 2013; Kawamura and Miura 2014a, b). By considering of a series of model test results, a prediction method for slope stability was proposed based on the water retention characteristics (e.g., the changes in water content) of slopes consist of volcanic soils. However, it has been highlighted that the results obtained from a series of small-scale model tests might change with variations in inherent errors, such as the scale effect.

Thereafter, the authors conducted field monitoring from September 2011 to July 2012 on a large-scale embankment that was constructed to 5 m in height, 2 m in crown width, 4 m × 3 (=12 m) in length, and at a 45° angle by using a representative volcanic soil in Hokkaido, Japan. Details of the results of the large-scale embankment were reported by Kawamura et al. (2013).

This study aims to identify the characteristics of a large-scale embankment reconstructed by a representative volcanic soil in cold regions during all seasons and to confirm the validity of the prediction method that has been proposed in previous studies. Taking into consideration the results of the field monitoring, model test, and stability analysis, the aspects of slope failure were clarified. In addition, it was found that rainfall-induced failure of embankments could be uniquely evaluated by the changes in water content in zones exposed to freezing and thawing. Finally, a stability monitoring protocol is discussed based on Horton’s (1939) infiltration theory in detail.

Figs. 7(a and b) show the slip lines of failed slopes and the shear strain distribution at failure for embankments constructed using Komaoka volcanic soil (Kawamura and Miura 2014a), respectively. For the case with a low water content of w₀ = 38% [Fig. 7(a)], it was shown that Slip line 1 (the first failure) was triggered around the toe of embankments, and then Slip line 2 (the final failure) rapidly developed with the increase of weight due to rainfall and of pore water pressure around the toe of the embankments, as shown in following sections. Therefore, the slope failed at 3,650 s.

In the case of a high water content of w₀ = 43%, surface flow due to erosion proceeded until the Slip line shown in Fig. 7(b), and the elapsed time reached the same depth as Slip line 2 for w₀ = 38%, which was 9,000 s. As shown in Fig. 3, the optimum water content of Komaoka volcanic soil was 40.5%. The failure mode differs for w₀ = 38% and 43% based on the developments of the saturation degree and in the excess pore water pressure (Figs. 8 and 9). For instance, pore water pressure for w₀ = 38% increased rapidly compared with that for w₀ = 43%. Due to the changes in the initial water content, a surface failure with a circular slip was caused for w₀ = 38%, and a surface flow failure was induced for w₀ = 43%. This was because the permeability of compacted soils generally decreased with an increase in water content and indicates the minimum value for higher water content over the optimum water content. Matsumura (2014) confirmed this reduction of permeability with increasing water content for compacted Komaoka volcanic soils, although it was presumed that the effect of particle breakage on permeability was reflected in the results. By considering the test results, slope failure appears to be induced by the expansion of areas with high water retention and the increase of pore water pressure around the toe of embankments.

A series of rainfall model tests with the freezing and thawing was conducted (e.g., Kawamura and Miura 2014a, b). During freezing and thawing, the slope surface of the model was frozen with dry ice over 8 h, and was thawed at a temperature of 20°C over 8 h after the model was set up; subsequently, a model test with rainfall continued until the model embankment failed. Representative changes in temperature (T1–T9) in a model embankment during freezing and thawing are shown in Fig. 10. This figure shows that the temperatures decreased during freezing action (until approximately 28,800 s), and then increased over time. Specifically, temperatures at the depth of 25 mm (T1, T4, and T7) indicated approximately < 0°C. A freezing depth of approximately 50 mm was visually observed in model tests.

Variations in the degree of saturation for model embankments composed of Komaoka volcanic soil are shown in Fig. 11. By comparing the cases without freezing and thawing (Fig. 8), no difference was observed in the behavior of saturation between both cases, but the difference in elapsed time to reach failure was observed. For instance, the elapsed time until failure for the case of freezing and thawing for w₀ = 43% was 765 s, which collapsed approximately 3.5 times faster than the case without freezing and thawing (2,690 s). To clarify the influence of freeze–thaw action on the deformation behavior of model embankments, Figs. 12(a and b) show the failed shape and deformation behavior (shear strain distributions) with freezing and thawing for w₀ = 43%. In comparison with cases not subjected to freezing and thawing [Figs. 7(b) and 12], the shape at the final slip line differed for each test case. For instance, the slip line depth with freezing and thawing was shallower than that without. The result means that hollows due to freezing and thawing produce loose structures in the frozen layer compared with before the freeze–thaw action. The difference in deformation behavior by the changes in the initial water content was not confirmed for Komaoka volcanic embankments with freezing and thawing. A similar tendency was observed for other volcanic soils (Kawamura and Miura 2013, 2014b).

To confirm this behavior, comparisons of the frozen layer (D_f) and the depths of the final slip line (D_s) measured in the model test are shown in Fig. 13. In this figure, test data was additionally plotted as reported by Kawamura and Miura (2014b) and illustrates that both depths were approximately the same for the model test, as shown by a dotted line. In addition, Kawamura and Miura (2013) clarified that rainfall-induced failure of unsaturated slopes formed from crushable particles was derived from loose structures generated by freezing and thawing and by the reduction of shearing resistance due to particle crushing during freezing and thawing. Of note, the influence of particle breakage on slope stability was small for Komaoka volcanic soil. Kawamura et al. (2013, 2016) indicated that model and large-scale embankments subjected to freezing and thawing for one cycle (one season) have already been in the plastic equilibrium. Specifically, normal displacements on the frozen surface were 5.4 mm (normal strain ɛ = 5.4%) for w₀ = 43% and 1 mm for w₀ = 37% (ɛ = 1.0%) in model tests. Converting these values into shear strain results in the range of γ = 1.5%–8.3%, which means that the embankment in high water content becomes the plastic equilibrium at one cycle of freeze–thaw actions. In addition, Matsumura et al. (2015) elucidated cyclic mechanical behavior of compacted Komaoka volcanic soils, and that cyclic strengths of specimens with a freeze–thaw sequence were lower than those with no freeze–thaw sequence. By taking into consideration the results, it was highlighted that areas of embankments affected by freezing and thawing for one cycle have already been in plastic equilibrium; therefore, estimation of an influenced area subjected to freezing and thawing is important for slope stability in cold regions.

In this study, a prediction method that considers the water retention characteristics for surface slope failure of such embankments was proposed. Fig. 14 shows the soil–water characteristics curve of Komaoka volcanic soil examined by test method for water retentivity of soils according to JGS 0150-2009 (JGS 2009b). In Fig. 14, a fitted curve was indicated by the van Genuchten model (1980). A test specimen was compacted at water content w = 38.2% and the degree of compaction D_c = 88.0% (Matsumura 2014). It was recognized that the specimen was drastically saturated at <10 kPa in suction. In the model tests, the initial water content of w₀ = 38%–43% corresponded to the degree of saturation of S_r = 53.4%–60.4%. The suction value (s) was <5 kPa when the water content was >w = 39.8%. Therefore, the contribution of suction on shear strength appears to be very small for Komaoka compacted soils. Due to these factors, the authors focused mainly on hydraulic mechanical behavior at failure for the following discussion.

It is well-known that Horton (1939) investigated infiltration phenomena in the ground during rainfall and proposed an empirical expression. In this model testing, a similar tendency where the degree of saturation increases with rainfall and converges at a certain value was observed (Fig. 8). For example, the following equation can be expressed by substituting the degree of saturation with water content based on Fig. 8 as follows:where w = water content (%) at time; w₀ = initial water content at time t = 0; w_c = final water content; and n = a constant for a given curve for the water content–time relationship. For example, the parameters w₀ and w_c, and n for the data of sm1 in Fig. 8(a) is 40.7% and 60.7%, and 0.001, respectively. Therefore, changes in water content, which could be easily measured, were available for stability evaluation.

To estimate water content at failure (w_f) the relationships between water content at the initial state (w₀) and the failure (w_f) (=final water content w_c) are shown in Fig. 15. In this figure, all test data are additionally depicted (including those from previous studies). The water retention capacity was evaluated as water content on the slip surface. From Fig. 15, it can be seen that there are unique relationships between both water contents. Their relationships can be estimated as follows:where β and γ = coefficients of a given curve whose values are shown in Fig. 15, where it can be seen that the water content at failure decreased for the case with freezing and thawing.

By examining the previous results, the authors conducted field monitoring from November 2012 to November 2013 for a large-scale embankment reconstructed to a height of 5 m, width 2.7 m, length 4 m, and an angle of 45° using the same soil as that of the small-scale model tests.

The field monitoring site was located in Sapporo, Japan, as shown in Fig. 1. A large-scale embankment was supported by lateral confining plates (the right and left sides), covered with plastic sheets (back side and bottom), and was thereafter built using a hand-guide roller (weight = 5.88 kN) by compacting Komaoka volcanic soil to achieve more than the degree of compaction (D_c) of 85% for each layer of 0.25 m. According to in situ density and water content tests, the compaction degree (on average) and water content (on average) were D_cave = 95.5% and w_ave = 42.5%, respectively. Details of the construction procedure are described by Matsumura (2014) and Kawamura et al. (2015).

The following instruments were installed to monitor soil behavior and temperature in the air and the embankment: (1) soil moisture meter [Time Domain Reflectometry type (TDR)], (2) tensiometer, (3) thermometer, (4) rainfall gauge, (5) snow gauge, (6) accelerometer, and (7) Anemovane, as shown in Fig. 16. The soil moisture meters and thermocouple sensors were set up at interval depths of 0.2 m until a depth of 1.5 m. The specifications of the instruments are reported in Kawamura et al. (2013). All data were collected within the sampling period of <10 min. In this study, volumetric water content (θ) from the TDR devices was converted by water content (w) by the relation w = (ρ_w/ρ_{d in situ}) θ, where ρ_w is the density of water and ρ_{d in situ} is the in situ dry density of soil respectively.

After the construction of the embankment, 10 pipes for water supply (Fig. 17) and 3 tanks were set in order to follow infiltration and freezing and thawing under climatic conditions and to cause slope failure. The tanks were located 3 m higher than the embankment crown, and the amount of water was controlled by each valve equipped with the tanks (Kawamura et al. 2013). The amount of water supply was approximately 250 L/day from November 20, 2012, to November 30, 2012. Thereafter, it was gradually increased to approximately 1,000–3,000 L/day (from May 7 to November 15, 2013). Water supply stopped for the winter months (from December 2012 to April 2013), and stability monitoring continued from November 1, 2012, to November 15, 2013. During monitoring, the effective responses of acceleration due to earthquakes were not observed. Therefore, the data from the accelerometers were omitted from this paper.

Fig. 18 shows variations in temperature in the embankment from November 1, 2012, to November 1, 2013. During monitoring, snow removal was carried out during the winter season (from December to March 2013). In Fig. 18, temperature for each depth varies daily; in particular, the value at the depth range of 0–0.2 m was <0°C from December 2012 to March 2013. Therefore, it was confirmed that the embankment at this location would be affected by freezing and thawing, because Komaoka volcanic soil contains a significant amount of fine particles, such as silt.

After the winter season, water supply resumed from May 7, 2013. During monitoring, surface failure with a large deformation was induced on October 17–18, 2013 due to the increase of rainfall and water supply. In subsequent investigations, the characteristics of the embankment slope before failure was investigated (Kawamura et al. 2015).

Figs. 19(a and b) show changes in the aspect of the slope surface with elapsed days based on surveying. For example, the toe of the slope was eroded on August 1, 2013, its area gradually extending proceeding slope failure due to the increase of water supply. Fig. 19(b) shows the slope shape after failure on October 19, 2013. According to field investigations, the depth of the failed area was approximately in the range of 0.6–0.9 m. The failed soil volume on November 15, 2013 was estimated to be approximately 7.5 m³.

Figs. 20(a and b) show changes in water content for SM_R2 and SM_R3 sensors in the failed area during monitoring (Fig. 19). It can be seen that the water content rapidly increased with the increase of rainfall and water supply, and decreased at slope failure; the depths of 0.2 and 0.4 m were particularly significant. Therefore, it was assumed that slip lines were generated from a depth of 0.2–0.4 m. Due to the development of slip lines, the failed area appeared to become deeper (to a depth of 0.6–0.9 m). Therefore, the change in water content is a key factor to evaluate slope stability.

Figs. 21(a and b) shows the behavior of pore water pressure measured by the T_C2 sensor until slope failure. This figure indicates that pore pressure was a negative value (−2 to −5 kPa) and was constant immediately before slope failure and that it sharply increased at failure; thereafter it decreased. This indicated that monitoring of pore water pressure using a tensiometer might be one method to predict surface failure for unsaturated slopes, similar to the change in water content. Details of data on other monitoring devices are described by Matsumura (2014).

Compared with model test results, the failed shape and the development of saturation degree indicated a similar tendency to that of embankments subjected to freezing and thawing. For instance, the scale (model/prototype) of length was 1/λ [=0.14(=0.7 m/5.0 m) in height] in this study. Based on the slip line depth of 51 mm that was obtained for the case of w₀ = 43% with the freezing and thawing (Fig. 13), the corresponding depth converted to this in the prototype (the large-scale embankment) was 365 mm. Therefore, the model test result explained the field data if the slip line was induced at the depth of 0.37 m in the field. In addition, the development of saturation was approximately the same compared with the corresponding depths [the sensor of sm3 at the depth of 25 mm (Fig. 11(b))] and the sensor of SM_R2 or SM_R3 at the depth of 0.2 m). However, transducers for pore water pressure could not be installed around the slip line under freeze–thaw action due to the capacity of sensors; however, it was presumed that the behavior of pore water pressure had a similar tendency as the large-scale embankment, although the difference in infiltration process of supply water was highlighted. However, it is recognized that the evaluation of the mechanical behavior of a large-scale embankment from the results of 1 g small-scale model tests is difficult in practice. In particular, because the corresponding points between a large-scale and a small-scale embankment were shallower zones in this study, the confining pressure effect on the mechanical behavior might be ignored. Further discussions are required in this area of research.

Based on the failure situation, the analysis was performed using the simplified Janbu method (Ugai 1987). Figs. 22(a–c) show a numerical model of the large-scale embankment at failure and models of representative distributions of water content on September 18 and October 17, 2013. In this study, the slope stability was evaluated by taking changes in water content (unit weight) into account, because the suction effect was very small when the water content was >w = 40% (S_r = 56%). The boundaries of both sides, the back side, and bottom were fixed and were undrained conditions.

Fig. 23 shows a variation in the safety factor (F_s) on the slip line with elapsed days. Numerical parameters adopted in this study are shown in this figure. It was found that the safety factor decreased daily and that it was <F_s = 1 on September 28, 2013. By considering the slope failure data, the embankment might have destabilized on September 28, 2013. The analytical results correspond well with the mechanical behavior at failure. Water content at the initial state (at the construction) (w₀) normalized by that at the failure (w_f, w₀/w_f ) was approximately 80% for this case. If approximately 20% of the increase in water content could be monitored for this case, stability evaluation should be performed.

Soil samples were collected from the zones on the slip line to investigate water content at failure. The water content on the slip line area was 54% on average. Based on the previous results for initial water content at construction and failure, the validity of Eq. (5) was confirmed (Fig. 15). It is noted that the field data of w = 54% (see star symbols) was approximately the failure line of freeze–thaw action for Komaoka volcanic soil. The result obtained from the field data might be expressed as an underestimated value, because there is a difference in depth between the slip line estimated by the behavior of soil moisture meters (=0.2–0.4 m) and the frozen depth (=0.2 m) estimated by thermocouple sensors as aforementioned; however, it was confirmed that the estimated value was in the range from no freeze–thaw action to freeze–thaw action in Eq. (5). Therefore, Eq. (5) adequately explains the field data for the embankment constructed from volcanic coarse-grained soil.

Therefore, the risk evaluation of embankments could be enabled by the proposed Eqs. (4) and (5) based on the changes in water content if the threshold of slope failure could be prescribed. For example, if the line of water content (w_f) potentially causing slope failure was estimated as shown in Fig. 15 and the depth of the frozen area (including frost heaving), the slope stability could be evaluated based on the change in water content measured by instruments, such as soil moisture meters.

Based on the field monitoring of a large-scale embankment, the model tests, and the numerical analysis conducted in this study, the following conclusions were obtained.

The authors wish to express their sincere gratitude to Ms. A. Kudo (Hokkaido University) and Mr. H. Nakagawa and Mr. K. Fukutsu (Muroran Institute of Technology) who conducted a major part of field monitoring, experiments and stability analysis, and Dr. S. Yokohama (Hokkaido University). This study was undertaken with the financial supports of KAKENHI [Grant-in-Aid for Science Research (A) No. 23241056, (C) No. 24560597 and (B) 20H02404], Japan Society for the Promotion Science.