The test soil is mainly composed of silty clay, which contains a small amount of gravel. The soil grain size distribution curve is shown in Fig. 3.
The saturated volume moisture content of the soil is 48.2%, and the residual volume moisture content is 6.0% using the Van-Genuchten (V-G) model for nonlinear fitting (Table 1).
Natural dry density (g/cm3) Saturated volume moisture content (m3/m3) Optimal mass moisture content (m3/m3) Maximum dry density (g/cm3) Liquid limit (%) Plastic limit (%) Shrinkage limit (%) Free swelling ratio (%) 1.59 0.482 0.165 1.71 35.7 18.4 8.2 42.5
Table 1. Physical parameters of the test sample
To be as consistent as possible with the physical properties of the original slope, the soil mass required for stratified volume was calculated in terms of the natural density of soil. After a period of drying, dispersing, and water spraying, the soil sample was filled in terms of the natural dry density of 1.59 g/cm3 and the mass moisture content of 13.8%. Subsequently, the loose soil in the flume was compacted to the layered design line by the vibratory plate compactor (i.e., 20 cm each layer). At the boundary of the model, the rubber hammer was used for supplementary compacting. The slope was filled in the form of steps, and the attendant excavated in terms of the slope designed after completion of filling. Finally, the slope would be placed for 48 hours to achieve the purpose of uniform water content in the slope.
Figure 4. The layout of the sensor and dynamic monitoring process. (a) The side view of sensor arrangement, six layers of TDRS, and four layers of piezometers are respectively arranged, and their positions are represented by Wxy and Fxy, where x represents the number of layers where the sensor is set, y represents the number of sensors in each layer. There are also five tensiometers and three FBG flexible displacement meters, whose positions are indicated by Tx and Dx respectively, where x is used to distinguish the position of the sensor. (b) The top view of sensor arrangement, showing the plane position of all sensors. (c) The spatial position coordinates of all the sensors, showing the spatial position information of the sensors. (d) Map of the mobile monitoring station showing the dynamic transmission process of all sensor data.
As can be seen in Fig. 4a, 18 Time-Domain Reflectometers (TDRs) were divided into six layers and installed inside slope for monitoring the volume moisture content at different locations. The piezometers (range 0–100 kPa, accuracy ±2 kPa) were 10 mm in length and 4 mm in diameter, which was installed on the same elevation as the TDRs with the aim of measuring the pore-water pressures. In order to measure the hydraulic responses at different locations of the slope, five tensiometers (diameter 22 mm; range 100 kPa; accuracy 2%) were installed vertically at the designated position along the slope. The measuring depth ranged from 30 to 70 cm (Fig. 4b). It is essential to ensure that bubbles in the tensiometers need to be removed in distilled water before installing the tensiometers (Kim et al., 2018). Meanwhile, considering the influence of the boundary effect, the sensor was not laid within 30 cm of the boundary.
The FBG flexible displacement meter was developed using the same method as Li et al. (2017). Three FBG flexible displacement meters were symmetrically distributed at 70 cm on both sides of the axial plane of the slope. The displacement gauge was made by coupling 1.4 and 2.8 m length polypropylene random (PP-R) tube with quartz fiber (28 cm spacing between fiber grating points) and was welding fixed between the support and the base (Fig. 5a). After embedding in the borehole, the surrounding of the displacement meter was filled with fine-grained soil for compaction to ensure that the displacement meter and the soil deformation could be coordinated (Fig. 5a, inset). Subsequently, the (FBG was connected to the M4151-16-25 high-performance demodulator to obtain the displacement.
Figure 5. Diagrams showing (a) the installation process of FBG displacement meter and (b) the testing principles and installation of horizontal measuring rope (inset).
Figure 5b is a schematic diagram of laying horizontal measuring rope after completing each layer of soil filling. According to Li et al. (2017) the FBG displacement meter could not measure the massive deformation displacement. Therefore, in the filling process, four horizontal measuring ropes, with an interval of 60 cm, were laid on each layer, and the front of the measuring rope was exposed to the slope. This aims to obtain the displacement of the slope during the failure by recording the change in the horizontal length of the measuring rope after each failure. The horizontal distance of each layer after failure was calculated using the following Eq. (1.1)
Here, Δxl, n represents the total horizontal displacement of the L layer at the nth deformation, Sl, n represents the sum of the horizontal displacements of layer L with n deformations.
The initial slope was restored based on a fixed point, and the three-dimensional (3D) spatial coordinates of each point of the slope were measured after sliding. Besides, the three-dimensional morphology of slope sliding failure was reconstructed by Rhinoceros software to calculate the volume changes of sliding bodies at different stages.
In this work, the artificial precipitation simulator was constructed with a height of 5 m, a length of 6 m and a width of 5 m. The height of the rain nozzles above the ground was 5 m, which is high enough for raindrops to reach terminal velocity (Kim et al., 2018). The simulator was fixed on the rail with a length of 12 m and moved together with the portable canopy. Three types of rotary down-flow nozzles (20 in total) were used as raindrop generators by opening in different combinations to achieve an artificial rainfall simulation process with different rainfall intensity. Subsequently, in the light of the calibration results, the intensity of light rain and heavy rain was set at 10.5–22.5, and 45.2 mm/h, respectively, to shorten the precipitation process. Before testing, the rainfall uniformity was tested with 100 measuring cylinders set within the range of 5 m×3 m to measure the rainfall volume, and the results indicate that the rainfall uniformity was more than 80%. Meanwhile, the filter paper test results show that the average diameter range of raindrop size varied from 1.473 to 2.562 mm.
Rainfall in the study area was primarily concentrated in spring and summer. In order to simulate the characteristics of long periods of light rain in spring and short periods of heavy rain in summer, the artificial precipitation was considered in three phases: continuous light rain, evaporation, and occasional heavy rainfall. In the first phase, a rainfall intensity of 17.5 mm/h was adopted to ensure that the slope will not produce immediate runoff and erosion. This phase of the rainfall lasted for 120 h. The second phase with a primary purpose of cracking the slope by evaporation lasted 72 h. In the last phase, rainfall intensity was 70 mm/h and lasted 48 h. The rapid re-wetting as a consequence of dry-wet cycles of the soil may also be attributed to the presence of desiccation cracks and macro-voids, which allow rainfall to preferentially infiltrate. It could promote the development of nearly saturated conditions near to the cracks and the macropores (Zhang et al., 2012; Greve et al., 2010; Römkens and Prasad, 2006), which allows us to evaluate the effects of the crack on the landslide development.
2.1. Soil Preparation and Slope Model Filling
2.2. Layout of the Instruments
2.2.1. Layout of the hydraulic sensors
2.2.2. Layout of the displacement meter
2.3. Precipitation Simulation
It has been revealed that the boundary effect could be ignored in the testing process (Lu et al., 2015; Jia et al., 2009). In this experiment, two gullies were formed at 25 cm on both sides of the boundary due to the concentration of runoff, and the erosion near the boundary. Fortunately, we did not lay sensors within the boundary of 30 cm, which provided a reliable guarantee for us to explore the slope failure mode accurately. The failure modes of the slope in different rainfall stages are shown in Fig. 6 and Table 2.
Figure 6. The failure mode of the model slope in different rainfall stages, showing the front view (L) and the top view (R) of the model.
Stage End time of deformation Form of deformation Deformation characteristics I continuous drizzle 26 h 27 min Erosion Two main gullies are formed near the boundary 62 h 15 min Scour deformation (forming gully) Six gullies with a depth of 2–5 cm are formed on the surface of the slope 100 h 08 min Creep damage The regressive slide appears in the middle of the slope II evaporation stage 122 h Creep of the slope body When the rainfall stops, the soil at the crest of the slope collapses 196 h Cracks deepened The fissure range is more extensive than the initial state, and the depth at the slope toe is 5–35 cm. Furthermore, the tensile fissure depth at the crest of the slope is 2–87 cm III heavy rainfall 208 h 32 min The whole slope collapses Surface runoff disappears, rainwater directly enters the crack from the slope crest, and then the whole slope collapses 240 h Creep and slide until stable With the continuous rainfall, the runoff mainly flows out of the slope along the gully. The trailing edge of the landslide collapses
Table 2. The whole process of slope failure under the influence of rainfall
After completing the slope filling, the moisture content of slope decreases under evaporation, and numerous shrinkage cracks occur at the crest and shoulder of the slope. These cracks significantly increase the hydraulic conductivity of the soil. During rainfall, the cracked soil around the shallow shrinkage cracks attains transient saturation, and the slip flow failure occurs in the cracked area (Figs. 6b, 6c). It is similar to the on-site slope failure phenomenon of Figs. 1a and 1e. However, such failure only occurs within 20–50 cm of the surface of the slope, and large-scale collapse is not observed during 120 h of rainfall.
In the evaporation stage, a large number of network desiccation cracks form at the crest of the slope (Fig. 6d). Continuous drying and creep tensioning of the slope lead to the gradual widening and deepening of these cracks. The runoff rapidly seeps along the wide cracks into the slope when heavy rainfall follows a period of evaporation. In this process, a large number of fine particles flow into the cracks, resulting in fissure healing (Fig. 6e). With the continuous heavy rainfall, a large amount of runoff is concentrated near the bottom of the deepest crack, causing the shear strength of the soil around the fissure to decline and eventually forming a continuous saturated layer that induced the slope instability.
In addition, it's noteworthy that there is a large amount of surface runoff on the slope when the slope is in the creeping stage. And the tensile crack on the slope crest is also enlarging. However, about 1 min before the overall slope instability (208 h 32 min), the surface runoff at the slope toe suddenly disappears, followed by a large-scale sliding failure (Fig. 6f). The reason for the phenomenon is supposed to be a continuous saturated zone, which is formed at the bottom of the crack when a large amount of runoff directly enters into the deep fissure from the slope crest (Wu et al., 2018; Fan et al., 2016).
The variations of moisture content and matric suction in different positions of the slope with time are at a shallow layer (Fig. 7a) and a deep layer (Fig. 7b). It is worth noting that previous studies have found that the moisture content at the slope toe responds faster, leading to traction failure of the slope (Wu et al., 2018, 2015; Fan et al., 2016; Zheng et al., 2009). However, in this work, the slope is firstly formed in the middle and upper parts, which is similar to the field slope failure (Fig. 1e). The reasons can be given as follows: (1) a large number of shrinkage cracks are formed on the slope crest during the evaporation period, and (2) slope shoulder is in the state of tensile stress, which is more liable to tensile cracks than the slope toe. Consequently, the permeability caused by these increased crack zones increases.
Figure 7. The variations of moisture content and matric suction with time in different locations. (a) Shallow layer, (b) deep layer.
Also, Fig. 7a indicates that the moisture content increases rapidly during the 0–48 h light rainfall stage due to the presence of a high matric potential (initial matric suction is 58.7–64.5 kPa). As the rainfall continues, the matric suction of the shallow layer gradually decreases. When the moisture content (W61) of the slope crest reaches 28.58%, the loss of fine particles leads to macropore enlargement in the soil, which breaks the original capillary water balance (Kim et al., 2018; Huang et al., 2009; Lourenço et al., 2006). Runoff continues to infiltrate along with the new pores, and the soil moisture content would be further increased (Lu et al., 2015; Lourenço et al., 2006). At this time, due to the increase of soil saturation after the wetting front, the rainwater infiltration rate decreases, and the growth rate of moisture content gradually decreases (W31 and W61). At 100 h 08 min, the moisture content of W31 in the middle of the slope is 27.39%, accompanied by deformation and collapse occurs in many places within 30 cm of the shallow layer. As the soil in the middle of the slope slides to the slope toe, the moisture content of W11 starts to increase significantly, soaring to 27.52%.
Figure 7b shows the variations of matric suction and moisture content with time at the deep layer. Due to infiltration hysteresis, the response of the moisture content in the deep layer is lower than that of the surface (Kim et al., 2018; Lourenço et al., 2006). The sensor (W53) located at a distance of 60 cm below the crest of the slope responds first, and at the end of 120 h, the moisture content continues to increase to 24.42%. And the moisture content increases by 30.3% compared with the initial moisture content, which can be related to the crack development at the crest of the slope and the high vertical infiltration rate. Besides that, the sensor (W33) in the middle of the slope (140 cm below the crest of the slope) has few changes due to the deep embedding. Also, since there are not many cracks in the initial stage of the slope toe, a large amount of rainwater results in slope runoff after the surface is saturated, which results in the moisture content at the slope toe increased by only 10.8% (W13).
On the other hand, Fig. 7 shows the results of the matric suction of the shallow layer at the crest (T5), the middle (T2 and T4), and the toe (T1) of the slope. According to Figs. 7a–7b, in the continuous drizzle stage (the end of 100 h), the matric suction of the crest, middle and toe of the slope is 7.9, 13.0, and 20.4 kPa, respectively, 87.6%, 79.7% and 65.1% lower than the initial value, which may be the main reason for the shallow slope failure.
According to Figs. 7a–7b, the moisture content of the slope in the evaporation stage generally decreases. In particular, the moisture content of the surface soil is more sharply dispersed. Compared with the moisture content at the end of 120 h, the average moisture content at and below the surface decreases by 16.4% and 9.92%, respectively. It may be due to the expansion, shrinkage, and creep of the soil surface, resulting in a more loose structure. At this point, the water and atmosphere exchange area increases, resulting in a higher evaporation rate (Seboong and Lu, 2015; Aleotti, 2004). Subsequently, in the evaporation stage, the free water between particles continuously evaporates, and the pores are gradually filled with air. This process resulted in a significant increase in the matric suction of the loose surface layer. As seen from Fig. 7, the tensiometer values of the shallow layer at the slope crest (i.e., T2, T4, and T5) increased to 13.3, 15.8, and 22.4 kPa, respectively. Moreover, due to the accumulation of water at the slope toe, the value of T1 only increased by 3.12 kPa.
Figure 7b shows that during the heavy rainfall after the evaporation stage, a sharp increase in the moisture content is detected at W33 at 208 h 32 min, which is installed at a distance of 140 cm under the flume surface. The moisture content of W33 reaches 41.8%, an increase of 205.7% over the initial moisture content. This phenomenon is attributed to preferential infiltration through the cracks (Zeng et al., 2018; Zhang et al., 2012; Römkens and Prasad, 2006). At this stage, the matric suction is at 0%, indicating that the soil is close to saturation, and causing overall collapse.
Previous studies indicated that the increase and dissipation of pore pressure could provide a better indication of failure (Wu et al., 2018; Lu et al., 2015). During the continuous rainfall (0–120 h), the pore water pressure in the surface layer fluctuates and increases (Fig. 8a), while pore water pressure in the deep layer can not increase significantly (Fig. 8b). It indicates that in the continuous drizzle stage, the slope deformation is dominated by shallow slide.
Figure 8. The variations of pore water pressure with time in different locations. (a) Shallow layer, (b) deep layer.
The pore water pressure fluctuation increases during the continuous rainfall mainly due to: (1) the uneven impact of raindrops cause the height of the overlying water column on the piezometers to fluctuate (Fan et al., 2016; Zhang M et al., 2015) repeatedly; (2) the erosion effect of the runoff leads to the loss of fine particles in the soil, and the pore water pressure decreases in the transient state (Kim et al., 2018; Greve et al., 2010; Huang et al., 2009; Lourenço et al., 2006; Römkens and Prasad, 2006). Due to the continuous rainfall stage, the rainfall intensity is even more significant than the soil hydraulic conductivity, and sufficient runoff quickly refills the pores, results in the continuously increasing of pore water pressure. As the pore water pressure continues to increase, the effective stress of soil is reduced. In Fig. 8a, the F31 sensor presents two sudden drop responses at 66 h 14 min, and 100 h 08 min negative pore water pressure decreases to -6.49 and -5.39 kPa, respectively. At this time, the slope surface is slip failure.
As shown in Figs. 8a and 8b, the pore water pressure on the surface layer dissipates significantly during the evaporation stage, while the pore water pressure in the deep layer does not decrease significantly. During the period of heavy rainfall after evaporation, the pore water pressure in different positions of the slope shows different characteristics. Due to the continuous rainfall and evaporation in the initial stage, the tension cracks and shrinkage cracks appear on the crest of the slope (Fig. 6d). They form an "intercepting ditch" at the slope crest, and the surface runoff infiltrates inside the slope through the cracks, forming a sudden increase of pore water pressure. For example, as shown in Fig. 8a, the pore water pressure (i.e., F31) rapidly increases to 10.17 kPa at 208 h 32 min.
On the other hand, the pore water pressure at the slope toe (i.e., F11) soars to 24.55 kPa under the continuous action of rainfall. Furthermore, the pore water pressure in the deep slope (i.e., F32) also increases to 17.7 kPa, indicating that the deep soil is already softened (Fig. 8b). Consequently, a rapid increase of pore water pressure reveals a reduction in effective stress and accordingly implies an impending slope failure (Lu et al., 2015).
The horizontal displacement changes of FBG displacement meter D1 (L=140 cm) at depths of 28, 56, 112, and 140 cm, respectively are shown in Fig. 9a, and the change of the cumulative horizontal displacement of the D1 FBG displacement meter is shown in Fig. 9b. In the process of light rainfall for 120 h, the displacement values of depth at 112 and 140 cm in the shallow slope gradually increase (Fig. 9a), which is attributed to the erosion deformation. It is consistent with the hydraulic response results mentioned above.
Figure 9. The horizontal displacement changes of FBG displacement meter D1 (L=140 cm). (a) Depths of 28, 56, 112, and 140 cm, respectively; and (b) the cumulative displacement change at 32, 67, 120, 132, 196, and 208 h.
Furthermore, in this work, the deep creep deformation of the slope during the evaporation stage is captured. According to Fig. 9b, between 120 and 132 h, the cumulative horizontal displacement of the slope increases by 48.3% to be 20.7 mm, compared with the displacement in the drizzle period. This phenomenon indicates that the deformation process of the cracked soil slope does not terminate with the cessation of rainfall during the evaporation period after rainfall. On the contrary, as the rainwater infiltrates into the deep soil, the shear strength of the soil gradually decreases with the loss of matric suction, thus leading to the continuous deformation of the slope.
Figure 10a shows the variation of the horizontal displacement of the FBG displacement meter D3 (L=280 cm) at depths of 28, 56, 140, 224, 252, and 140 cm, respectively. And Fig. 10b shows the changes of the cumulative horizontal displacement of the D3 FBG displacement meter. It can be observed from Fig. 10a that the D3 sensor changes little in the range of 140–280 cm below slope crest in the light rain stage (0–120 h), indicating that soil has no horizontal deformation. During the evaporation period of 120–130 h, the displacement also increases slightly, which is consistent with the response of the D1 sensor. Figure 10a shows that the maximum cumulative displacement of the shallow layer is 1.848 cm in the continuous light rain stage during the first 120 h. When heavy rainfall follows a period of evaporation, the surface displacement increases to 4.31 cm. Along with the continuous rainfall, the slope collapses after 208 h when the maximum horizontal displacement exceeds 5.07 cm.
Figure 10. The horizontal displacement changes of FBG displacement meter D3 (L=280 cm). (a) Depths of 28, 56, 140, 224, 252, and 280 cm, respectively; and (b) the cumulative displacement change at 32, 67, 120, 132, 196, and 208 h.
Subsequently, the massive deformation displacement is also measured according to the length of the measuring rope. According to Fig. 11a, at the end of 240 h, the maximum cumulative horizontal displacement at the slope crest reaches 147.7 cm, and the displacement at the distance of 160 cm below the slope crest reaches 133.3 cm, indicating that the deep soil moved during the heavy rainfall period. Figure 11b shows the three-dimensional form of slope failure. At 240 h, the maximum horizontal displacement of the slope crest increases by 94.5 cm compared with 208 h. At the end of the shallow landslide (end of 208 h), the volume of the slope decreases to 34.08 m3 compared with the initial state (36.72 m3). Nevertheless, a heavy rainfall follows a period of evaporation (end of 240 h). The volume of the slope only decreases to 25.03 m3, which is 31.8% lower than the initial state. It can be shown that a slope with cracks is far more destructive than a slope without cracks.