3.2 Field investigation
(a) Statistics from 2002 to 2008 (Fig. 7) indicate that the erosion rates in Sect. 3 − 1 and 3 − 2 were both greater than 10. In contrast, the erosion rates in the other sections were less than 3. The statistics from 2008 to 2016 displayed lower erosion rates and the approach of stability. The cliff lines of Pit No. 4 in the remote sensing images from the three years showed less significant slope retreat in Pit No. 4. The most significant cause of varying degrees of erosion was the uneven distribution of erosion gully development. In the study area, the erosion gullies in Pit No. 3 were the most severe.,
(b) Based on the field investigation, the landslide mechanisms in the source area could be divided into five stages. In the first stage before rainfall, the gravel and fine particles reach equilibrium, which promote slope stability. In the second stage, when rain begins to fall and the surface runoff increases, the gravel becomes wet due to its high porosity. In the third stage, continuous rainfall leads to the continuous scour and loss of fine surface particles, which do not immediately fall into the stream but are deposited at the slope toe. The larger gravel remaining on the surface support one another, and the slope begins to lose stability. In the fourth state, the rainfall increases, leading to the excessive loss of fine particles and the gradual formation of the sliding face. Due to rising pore pressure and reduced shear strength in the gravel slope, the gravel layer begins to slide. In the fifth stage, after the landslide, tension cracks form at the top of the gravel slope, and the cycle repeats.
(C) In potential landslide stability classification (Fig. 8), the slope stability classification proposed by Crozer (1984) indicates that the study area belongs to Class I, unstable slopes. There is evidence of significant changes in recent years. Ia indicates highly active landslide slopes on which the landslide materials are still sliding at a total of 17 places. Ten of these places are in Pit No. 10, where the hazardous sliding masses are all at the source. Ib marks the landslide materials that are still sliding, with six places showing signs of sliding further. Ic are the slopes on which landslides have taken place in recent years. These slopes show evidence of landslides occurring in recent years at 11 places.
3.3 Numerical simulation of landslides
The grey region marks the study area, and the red balls are the historical landslide areas. Having the high-precision aerial photos from 2008 and 2016, we could perform simulations and compare the results with the onsite conditions. Images from Formosat-2 for 10 years provided more information for verification. Figure 9 shows the simulation of the historical landslide processes in 2008. The simulation results were as follows:
(a) With ball element radius set at 2.5 m and 5,380 ball elements in total, the depth of the historical landslide in Pit No. 1 ranged from 9 m to 13 m. The ball elements were arranged in three layers. The depth of the historical landslide in Pit No. 2 ranged from 8 m to 11 m, the ball elements also arranged in three layers. The maximum depth of the historical landslide in Pit No. 3 reached 26 m, and the ball elements were arranged in six layers. The depth of the historical landslide in Pit No. 4 was approximately 7 m, and the ball elements were arranged in two layers. The depth of the historical landslide in Pit No. 5 was roughly 12 m, and the ball elements were arranged in three layers.
(b) At Step 9,400, the sliding speed in Pit No. 5 was higher than those in the other four pits. The river channel in Pit No. 5 is straighter than those in the other four pits, and the source area was the lowest in elevation. At Step 294,000, most of the sliding mass in Pit No. 3 in the image had already been deposited in the stream, while those in the other pits were gradually being deposited at the river mouth. At Step 774,070, the gravel in Pit No. 3 was moving more slowly in the middle section of the river channel, which was bent slightly and somewhat obstructed the rolling of the gravel. At Step 1,254,070, the majority of the gravel had left the river channel and entered the alluvial fan below.
(c) The gravel transport and alluvial fan deposits in the remote sensing images from 2008 and 2007 were similar to those in the numerical simulation at Step 554,070, which indicate no significant errors in the parameter settings. The debris washed down in Fig. 9 did not pile upwards but moved towards the sides. Only when the debris encountered obstacles in the terrain did is continue to pile upwards. The transport conditions in Pits No. 1 and 5 in the numerical simulations showed that without disaster prevention measures, typhoons would bring down substantial amounts of debris, thereby burying the tunnel exits. The tunnel also had to bear the weight of the debris washed down in Pit No. 3, which means that regular tunnel checks are needed to preserve the safety of road users. Figure 10 displays the simulation results of the historical landslide processes prior to 2016:
(d) With ball element radius set at 2.5 m and 7,380 ball elements in total, the layer arrangements in Pits No. 1 through 5 were the same as those for the 2008 simulation. At Step 0, the ball elements at the top in Pit No. 3 had already passed the ridgeline. The landslide area was greater than that in 2008, while the remaining landslide areas were mainly in the source areas of the respective pits.
(e) As the landslide area had already passed the ridgeline, some of the ball elements in the source area of Pit No. 3 fell off the sides at Step 194,050. The remainder of the sliding mass fell in the river channel. At Step 674,051, the alluvial fan gradually formed. Observation revealed that the deposits on the alluvial fan in Pit No. 3 at Step 674,050 were different from those in the 2008 simulation. Originally, the deposits continuously extended to the sides and rarely piled upwards. However, in this simulation, the deposits piled upwards, possibly due to the larger landslide area in Pit No. 3 as well as the significantly larger landslide area on both sides of the river channel at the middle section. This resulted in debris being washed down from the source area after the landslide in the middle section, which caused a substantial amount of debris being moved to the alluvial fan at once.
(f) A comparison of the numerical simulation with the remote sensing image of 2016 revealed that most of the alluvial fan had been covered in vegetation by 2015. The deposits on the left of river mouth No. 3 were higher than those on the right. The debris scoured down moved downwards. In 2016, river mouth No. 3 showed signs of debris flow movement, and once the alluvial fan deposits above the tunnel reached a certain height, they extended towards the riverbed.