Four main types of methods were involved in this study including detailed geological surveying, deformation monitoring, data analysis and stability analysis. To identify landslide characteristics, the site was visited for field survey, and the geological features were mapped on a 1 : 1 000 scale geomorphological map. A laser rangefinder was used to precisely map the locations of ground cracks on the landslide, and the cracks were then put on the geomorphological map. Sixteen boreholes, with depths ranging from 43.5 to 68.1 m, were drilled and cored to identify the material, component, thickness and the slip surface of the landslide. To better understand the landslide structure, profile 1-1' was sketched to interpret the layers and drilling work. Two adits perpendicular to the Beier Tunnel were set to find any other latent slip surfaces and determine the spatial location-relation between the slip surface and the tunnel. The sliding soil was sampled in AD1, and the physical and mechanical parameters were obtained by laboratory test conducted.
Ground-based network has been established since Dec., 2011, the monitoring network took in 8 surface optical targets on the landslide. The landslide surface displacement was monitored by ground-based optical targets using electronic total station, with rectangular coordinate system: axis Y is the sliding direction, and axis H is the gravity direction.
Reservoir levels and local precipitation data were obtained from the PowerChina Chengdu Engineering Corporation Limited, and China Meteorological Data Service Center (CMDC), respectively. The statistical significance of the correlation between time series of reservoir fluctuation and landslide displacement was calculated with the cross-correlation function (CCF) of Statistical Product and Service Solutions (SPSS) to identify the lags of the landslide displacement that might be useful for establishing reservoir regulations. The CCF between the reference time sequence x(t+h) and the time-shifted sequence y(t) is defined as
where CCF(h) is the averaged product of y(t) lagged with respect to x(t). The value of CCF(h) ranges from -1 to 1, and the value of zero states that there is no correlation between the two compared time-series. When the absolute value of the cross correlation is high for some value of the lag h, it can be said that x(t) and y(t) are similar at this lag value (Roth et al., 1971).
An associated GeoStudio analysis was used to evaluate the landslide stability under established reservoir regulation. Based on the schematic model of Xierguazi Landslide from the 1-1' geological profile, the numerical model is set up for simulation. Module seep was run to simulate the seepage flow during the water level fluctuations of pre-flood drawdown control and flood season filling control. Furthermore, the pore-water pressure condition from seep simulation was applied on the following stability analysis of module slope. Physical and mechanical parameters of the materials for GeoStudio are listed in Table 1.
Materials Permeability coefficient
Poisson ratio Cohesion
Internal friction angle
Deposits 6.35×10-4 8.1 0.3 75 33 2.28 2.1 Sliding soil 1.33×10-5 1.2 0.35 58 24.7 1.85 0.1 Bedrock 5.787×10-7 18.5 0.22 150 38 2.55 3.5
Table 1. Physical and mechanical parameters for GeoStudio model
No historic landslide activity in this area was recorded previously, but the geomorphic evidences found suggest that the Xierguazi Landslide is an ancient landslide, and the movement observed in Sept. 2011 was a reactivation of the lower part of the dormant accumulation. As seen in Fig. 2c, a drastic change in slope occurs from 2 602 to 2 644 m in elevation on the cliff. Whereas the rest of the slope is around 30°–40°, this 40 m region is a steep cliff of 50°–55°, with a 30 m wide platform directly underneath it. Although no cracks nor scars were found in the vicinity of the cliff due to the long-term weathering process, the special landform is deemed an evidence of landslide, and the cliff should be considered as the head scarp of the ancient landslide.
The geometry of the Xierguazi Landslide is delimited by two continuous narrow grooves on the slope, a front edge submerged by impounded water, and an arc-shaped cliff as the back edge (see Fig. 4). It is an oval plane with a maximum length of 1 057 m, a maximum width of 450 m, and an area of approximately 0.43 km2.
According to the cores from the drilling work, the landslide accumulation can be divided into three layers of materials (Fig. 5). A very thin layer of fine-grained loess from eolian deposits through time presents on the surface. The loess is distributed not across the entire slope but only at higher elevations. The second layer, covered by the loess layer, is gravel soil with a thickness varying from 10 to 25 m. The main contents of this layer are 65% metasandstone gravel, 20% breccia particle and 15% sandy-clay. The bottom layer is sandstone blocks reaching almost 27 m in thickness, and is mainly distributed in the lower part of the accumulation. Numerous drilled core columns that are more than 20 cm long reveal that this layer is composed of a huge number of integrated sandstone blocks at depths from 20 to 50 m.
Figure 5. 1-1' geological profile of Xierguazi Landslide, see Fig. 4 for the profile location.
As shown in Fig. 4, the major sliding direction of ancient landslide generally pointed toward the Heishui River and the ground deformation primarily featured long and big tension cracks. According to the geomorphological features and deformation characteristics of the accumulation, the ancient landslide can be classified into 4 sub-zones as below.
Zone A (including A1, A2 and A3) is the reactivated zone of the ancient landslide below the elevation of 2 320 m that formed after the impoundment, and each of the sub-zone has different deformation behaviors.
Zone A1, intense deformation zone. Masses of big and long cracks are interwoven in the middle part of this zone, mostly having an extending direction from SE 30° to 40°, lengths of 3 to 10 m, and widths of 5 to 20 cm. The head scarp of this zone (see Fig. 2d), featured multiple scars, and implied that there were multiple intermittent landslide movements. From those scars, we concluded that Zone A1 had at least undergone 5 movements with displacement of 1.1, 0.4, 0.3, 0.6, 0.5 m, respectively.
Zone A2, long sliding-distance zone. This zone slides down a maximum vertical distance of 21.7 m as we measured from the head scarp. The blocks retain the features of their original strata, with two groups of tectonic joints measured. A 60 m long crack at the elevation of 2 210 m is observed; the crack opens an average width of 1.4 m with maximum visible depth of 2.8 m, and it reveals that Zone A2 had undergone a secondary slide. When the secondary slide ran faster than the main slide, the crack was formed in the middle of this zone.
Zone A3, reservoir bank reformation zone. No long and big crack was found in this zone, except the trebling bank surface collapses in the front of the slope. The bank surface collapse is the outcome of reservoir fluctuations, which is known as reservoir bank reformation.
Zone B is a overburden creep zone. Unlike Zone A, there are only 6 long and big cracks on the slope surface of this zone. The biggest one is at an elevation of 2 450 m with a length of 150 m, and the others are about 10–30 m long. Judging from the mosses growing in the cracks, these large-scale deformations, which are deemed creeps, have been present for at least 3 years.
Based on the drilling work, a 0.5–1.5 m thick shear zone was located at a depth from 41 to 54.5 m, and registered a landslide accumulation with total volume of 1.35×107 m3. Moreover, we were able to collect the sliding soil at a horizontal distance of 38.5 m from the entrance of the adit AD1. The soil is tightly compacted, and is composed of 55% pale yellow clay, 30% fine sand and 15% angular gravel. The spatial relation between the landslide and the Beier Tunnel was confirmed as well, that is, the Beier Tunnel sits underneath the bottom of landslide with a vertical distance of 21 m.
Based on the measured surface displacements on the landslide, the average sliding direction of the whole accumulation was N40°E, and the displacement vector is shown in Fig. 4.
The time series of cumulative vertical displacements (Axis Reservoir Regulation for Control of an Ancient Landslide Reactivated by Water Level Fluctuations in Heishui River, China H in Fig. 6a) for surface points on the landslide clearly show that the deformation varied spatially, as displacements on Zone A are much greater than that on Zone B. From Dec. 9, 2011 to Sept. 15, 2012, the maximum displacement in Zone A exceeded -300 mm by which was measured on point DM1, and the minimum one even reached -132.5 mm at point DM6. Contrarily, the maximum displacement in Zone B was only -69.6 mm where the monitoring point DM7 was positioned, that suggests the reactivated zone (Zone A) moved faster and farther than the rear part of the ancient landslide. In addition, the displacements observed on different sub-zones of Zone A had distinct features, namely, maximum displacement in Zone A1 was greater than that in Zone A2, while Zone A2 displaced farther than Zone A3. It is also interesting to note that the curves for DM3 express a sudden change in vertical displacements, meaning that blocks in the vicinity of point DM3 experienced a surface collapse on Mar. 6, 2012.
Figure 6. Diagrams showing the cumulative vertical surface displacements. (a) Monitoring period before the operation of reservoir regulation, all the data were well collected except the period of Jan. 17–29, 2012 due to Chinese New Year; (b) monitoring period after the operation of reservoir regulation.
Fluctuation of the reservoir level appears to be the dominant factor influencing displacement of the Xierguazi Landslide. Visually, the cumulative displacement curves suggest that displacements of Zone A (DM1 to DM6) correlated with the reservoir level better than those of Zone B (DM7 to DM8). The reactivated zone (Zone A) deformed rather slowly during the filling stage from Apr. 14 to Aug. 4, 2012, yet quite clearly accelerated during the drawdown stage from Feb. 3 to Apr. 13, 2012. The vertical displacement on reactivated zone registered average rates of -0.6 mm/day during the filling stage, while the mean vertical displacement rates of the monitoring points in Zone A increased to -1.3 mm/day during the drawdown stage. The deformation in Zone B seems to be characterized by overburden creep during the entire monitoring period, for the cumulative displacements neither corresponded with the filling stage nor the drawdown stage, but increased with time gradually.
Surface displacement monitoring for 10 months in period 1 showed that the displacement of Xierguazi Landslide may be controlled by combined factors, especially by events such as the reservoir level fluctuations and the short-time severe rainfall. To examine the relationship between the changes of reservoir level and rate of landslide surface displacement during the monitoring period, we first divided the monitoring period into drawdown and filling stages (see Fig. 6a). Then, numerous statistical connection analyses were made to address the best correlation coefficients between the reservoir level fluctuations and the distinct landslide displacement datasets. Ultimately, we found that there is a significantly lag correlation between reservoir level change and surface displacement rate in Zone A during the drawdown and filling stages, separately.
The reservoir level change and surface displacement rate are both time series with units presented as m/day and mm/day, respectively. Their lead-lag correlation during drawdown stage is shown in Fig. 7a. Obviously, the correlation for surface displacement rate lagging behind the reservoir level change by 1 to 2 days is statistically significant from DM1 to DM6. More specifically, the best correlation shows a lag of 1 day is characterized in DM1 (r=0.915), DM2 (r=0.792), DM4 (r=0.828) and DM5 (r=0.724), while 0 day is lagged in DM3 (r=0.472) and 2 days are lagged in DM6 (r=0.813).
Figure 7. Diagrams showing lead-lag correlations between the reservoir level change and surface vertical displacement rate during drawdown stage (Fig. 7a) and filling stage (Fig. 7b). A positive value for lag number is a correlation between the reservoir fluctuation variable at a time before t and the landslide displacement variable at time t, it is sometimes said that reservoir fluctuations lead to landslide displacements. A coefficient value exceeding the upper coefficient limit or lower coefficient limit suggests strong correlation between two variables.
A significant negative lagged correlation is observed during filling stage as shown in Fig. 7b, which indicates that the reservoir level change is correlated with the surface displacement rate by 3 to 5 days earlier from DM1 to DM6. To be specific, a lag of 4 days is characterized in DM1 (r= -0.701), DM2 (r= -0.758), DM3 (r= -0.649) and DM5 (r= -0.725), whereas 3 days are lagged in DM4 (r= -0.438) and 5 days are lagged in DM6 (r= -0.563).
Previous studies have demonstrated that to maintain stability in reservoir regions, one must minimize the impact of water by managing variation in reservoir levels (He et al., 2008). To set up effective reservoir regulations, we should first confirm: (1) the length of time we should maintain a fixed reservoir level after substantial level changes, and (2) the maximum rate of level change we can allow at different intervals of reservoir elevation. The former can help examine the subsequent displacement of the landslide to ensure its stability, and the latter is to avoid any acceleration to the landslide that exceeds the maximum allowable levels.
By using the lead-lag correlation, the two points above for reservoir regulation are obtained. Firstly, the surface displacement rate lags behind the reservoir level change by 1 to 2 days during the drawdown stage, and by 3 to 5 days during the filling stage. To be on the safe side, 2 days for the drawdown stage and 5 days for the filling stage are set to maintain a reservoir level after a substantial level change.
Secondly, according to the time series data, it is worthwhile to note that the landslide displacement is controlled not only by the level change rate, but also by the reservoir elevation. For example, when the reservoir level raised to 2 125.5 from 2 124.5 m a.s.l. with an average rate of 0.5 m/d, a mean displacement rate of -1.9 mm/day was registered with the lag of four days; whereas a mean displacement rate of -1.3 mm/day was correlated with an average reservoir level change of 1.6 m/day four days before, when the reservoir level raised from 2 081 to 2 091 m a.s.l. in 6 days. This indicates that when the reservoir elevation is high, the landslide displaced more following the reservoir impounding than that when the reservoir elevation is low. This is similar in the reservoir drawdown stage. To keep the landslide displacement rate at an acceptable speed, we selected maximum rates of change in reservoir levels that can be allowed based on the stage of change (impounding or drawdown) and the current reservoir elevation. By using GeoStudio, plenty of trails from our landslide stability numerical analysis were conducted to find the best rate of reservoir level change which would not trigger a sudden failure of the landslide under different scenarios. To make sure the landslide would not fail, the factor of stability should be greater than 1 at any transient time during the reservoir level drawdown stage and filling stage. From the stability analyses, we concluded that: (1) the reservoir level drawdown rate should be -0.5–0 m/day when the reservoir elevation is above 2 083 m a.s.l. and be -1–0 m/day when the elevation is below 2 083 m a.s.l., and (2) the reservoir level impounding rate should be 0–0.5 m/day when the reservoir elevation is above 2 113 m a.s.l. and be 0–1 m/day when the elevation is below 2 113 m a.s.l.
Thus, based on our results as well as previously identified factors, we established a reservoir regulation that takes multiple factors into account (see Table 2).
Pre-flood drawdown control Flood season filling control Reservoir elevation (m a.s.l.) Level change regulation Fs Reservoir elevation (m a.s.l.) Level change regulation Fs When > 2 103 -0.5–0 m/day 1.02 From 2 063 to 2 093 0–1 m/day 1.11 When at 2 103 Keep the level for > 2 days 1.06 When at 2 093 Keep the level for > 5 days 1.07 From 2 103 to 2 083 -0.5–0 m/day 1.02 From 2 093 to 2 113 0–1 m/day 1.09 When at 2 083 Keep the level for > 2 days 1.08 When at 2 113 Keep the level for > 5 days 1.05 From 2 083 to 2 063 -1–0 m/day 1.05 When > 2 133 0–0.5 m/day 1.04
Table 2. Reservoir level regulation of Maoergai Reservoir, the Fs represents the lowest factor of stability of the landslide during different scenarios simulated by the numerical model