1 Introduction

Natural dams can be generated as a result of climate factors and/or geological processes. Typically, natural dams include landslide dams, which may be generated by earthquakes or heavy rainfall events. The moraine dams of glacier lakes can also be considered as natural dams. Natural dams are usually composed of loose material, hardly consolidated, and may fail by the overtopping flow or through seepage, which can lead to catastrophic flooding in the downstream. Flooding due to landslide dam failure has occurred worldwide (e.g., Costa and Schuster 1988; Cruden et al. 1993, 1997; Bovis and Jakob 2000; Korup 2002; Miller and Cruden 2002; Nicoletti and Parise 2002; Abril and Knight 2004; Dai et al. 2005; Korup and Tweed 2007). Typical examples include the October 1999 case in Poerua River in Westland, South Island, New Zealand (Davies et al. 2007) and the June 1967 case in the Yalong River in China (Li et al. 1986; Chen et al. 1992). Most recently, on May 12, 2008, a major earthquake occurred in Sichuan Province in China, which led to the formation of a series of major landslide dams (e.g., Hu et al. 2009; Cui et al. 2009). One of the largest, the Tangjiashan, landslide dam might fail due to the continuing inflow from the upstream, which could trigger catastrophic flooding in the downstream (the risk was later negated by an artificial spillway channel through the dam). Also, global warming is known to cause glacier retreat, and glacial lakes are formed or expanded, and many glacial-lake outburst flood events have occurred in history (e.g., O’Connor and Baker 1992; Cenderelli and Wohl 2001, 2003; Liu et al. 2003; Shen et al. 2004; Balmfroth et al. 2008, 2009).

A clear understanding of the flooding due to landslide dam failure is certainly necessary for disaster prevention and mitigation. Since real-time in situ measurements of the floods are difficult to make, experiments have been carried out in laboratories, which help enhance the understanding of dam-break hydraulics. However, previous physical experiments are constrained by the comparatively small spatial scales (e.g., Coleman et al. 2002; Chinnarasri et al. 2003; Zhu et al. 2006; Balmfroth et al. 2008, 2009; Schmocker and Hager 2009) and thus may not be able to fully reveal the complicated mechanism of the flood flow. In addition, previous experimental studies have mostly focused on the dam failure process, while the flood hydrographs are sparsely observed. For example, Coleman et al. (2002) only measured the outflow discharge of the breach, but ignored the flow conditions upstream and downstream of the dam. Chinnarasri et al. (2003), Zhu et al. (2006) and Balmfroth et al. (2008, 2009) did not measure the flow, but the geometry of the failing dam. Indeed, a range of factors involving dam structure and composition as well as hydrodynamic actions may affect the rate and size of breach formation (Morris 2009). In this connection, the experiments by Coleman et al. (2002) and Chinnarasri et al. (2003) are limited to the failure of the dams composed of purely non-cohesive sand under the action of a single inflow discharge. It is certainly critical to understand the influences of the inflow discharge, dam height, and composition on the dam failure and flood processes. Carrivick (2010) tried to conduct experiments of dam break flood over mobile bed in a flume 4 m × 0.5 m × 0.2 m. Yet, there are critical issues open to questions from hydrodynamic perspectives. First, the flow generated by partly raising the lock gate is not dam break flood, rather it is flow under sluice gate, which has been widely investigated in the context of hydraulic engineering. Second, rather coarse sediments (with medium diameters 5, 10 and 15 mm) are used in the experiments, while the flow is very shallow (3–8 cm). The flow depth-to-bed sediment size is far from the real cases in the field, leading to substantially twisted flow structure compared with the field flood. More critically, similitude of sediment movement in the laboratory model and prototype is not considered at all. While field experiments are conducted (e.g., Hoeg 1998; Hoeg et al. 2004), observations of the flood processes are largely hampered because of the risk to damaging the instrumentation and to the personnel and accordingly are normally insufficient for unraveling the mechanism of the dam failure processes. Succinctly, the understanding of the mechanism of landslide dam failure and the flood has so far remained poor and observational data sparse, which constitutes one of the basic impediments to the calibration and testing of mathematical models for the phenomenon.

This paper presents a series of experiments to study landslide dam failure and the flood. The experiments are carried out in a flume of 80 m × 1.2 m × 0.8 m. An array of twelve automatic water-level probes is set at the center of each cross-section along the flume to record the transient stages. Compared with previous experiments on small-scale facilities (e.g., Coleman et al. 2002; Chinnarasri et al. 2003; Zhu et al. 2006; Balmfroth et al. 2008, 2009; Schmocker and Hager 2009), the present experiments cover a wide range of inflow discharges from the upstream, initial dam height, and initial breach dimension. The experimental dams are composed of non-cohesive sand, sand–clay mixture, or sand–gravel mixture. The stage hydrographs and video recordings of the dam failure processes for each run of the experiments are obtained. The data help reveal the mechanisms of dam failure and flood propagation. Use of the experimental data is exemplified in the companion paper on shallow water hydrodynamic modeling of landslide dam failure and flood propagation (Cao et al. 2011). It is noted that dam-break flood, in general, may trigger active sediment transport (debris flow) and bed deformation, which, however, has been addressed elsewhere (e.g., Capart and Young 1998; Cao et al. 2004). Therefore, the present experiments feature a fixed bed initially (i.e., there is no sediment over the channel bed before the onset of the dam failure), and there is no sediment input from the far upstream end. As such, all the sediment involved in the experiments is from the failure of the dam.

2 Experimental setup

The experiments are carried out in a flume, 80 m long, 1.2 m wide, and 0.8 m deep. The flume is made of glass with a fixed bottom slope of 0.001, and the Manning roughness is approximately 0.012. The experimental setup is illustrated in Fig. 1. The dam is located at 41 m from the inlet of the flume. Twelve automatic water-level probes are mounted at the center of twelve cross-sections (CS1 through CS12), while CS1 is located 19 m from the inlet of the flume. The automatic water-level probes can track transient stages. The tracking speed of the water-level probes is set at 100 mm/s, and the corresponding measurement error is within ±0.5 mm. The sampling frequency is set at 2 Hz. A computer is deployed to process the stage data transmitted from the water-level probes. To facilitate the initialization of the operation of the automatic water-level probes, a 0.15-m-high weir is fixed at the outlet of the flume to keep a small depth of water in the downstream of the dam under the initial condition, and in the upstream of the dam, there is also a similar device around the inlet of the flume to hold some water under the initial condition.

Fig. 1
figure 1

Experimental setup. Twelve automatic water-level probes are mounted at the center of the flume to track transient stages

Full size image

3 Experimental procedure and data

The schematic diagram of the experiments is illustrated in Fig. 2, along with hydraulic and geometric parameters. Prior to the start of a run of experiment, there is some static water up- and downstream the dam, which is necessary for initializing the operation of the water-level probes. As an inflow discharge is facilitated through the inlet of the flume, the water level upstream the dam gradually increases, and once the flow overtops the dam crest, dam failure commences through erosion, i.e., overtopping erosion (dam failure due to seepage is not considered in this study). The initial static water depths immediately upstream and downstream the dam are H u and H d: Q is inflow discharge from the upstream; W is the initial width of the dam crest (=20 cm); and H is the initial dam height. In the present work, two different types of dams are considered, respectively, i.e., dams without and with initial breach. For the dam without initial breach, the dam height is uniform across the flume, and the longitudinal profile is illustrated in Fig. 2(a). The initial upstream and downstream slopes of the dam are denoted by S 1 and S 2, respectively. For the dam with initial breach, the initial dam height H is a fixed at 40 cm, a trapezoidal breach is initially set at the center of the dam, of which the depth is H b, the base width is 4 cm, and the side slope is fixed at 1:1, as illustrated in Fig. 2 (b). Except the initial breach, all the other geometrical conditions of the dam are the same as the dam without initial breach.

Fig. 2
figure 2

Schematic diagram of dam geometry and hydraulic parameters: a longitudinal profile of the dam and b dimension of initial breach

Full size image

Natural landslide dams can be composed of a wide range of sediment sizes, from clay, sand, gravel, and cobble to boulder (Costa and Schuster 1988; Korup 2002). In principle, for hydraulic similitude between the laboratory experiments and the prototype (field) phenomenon, the sediment for the experiments needs to be downsized from that in the field. In the present experiments, the dams are composed of purely non-cohesive sand (SA), a mixture of the sand and cohesive clay (SC) or a mixture of the sand and gravel (SG). The gradation curves of the SA, SC and SG are shown in Fig. 3. The median diameters are 0.8, 0.7, and 1.7 mm, respectively, for the SA, SC, and SG samples, which have a common repose angle of 30° approximately. For the sand–clay mixture (SC), the volume ratio of the clay to the total sand–clay mixture is 25%. While the composition of natural landslide dams can be heterogeneous, it is hard to find a general description of the heterogeneity. Thus, the present experiments are limited to homogeneous dams.

Fig. 3
figure 3

Gradation curves of dam materials, including non-cohesive sand (SA), sand and cohesive clay (SC), and sand–gravel mixture (SG)

Full size image

In total, sixteen runs of experiments are conducted for the failure of dams without initial breach (F-Cases) and twelve runs with initial breach (S-Cases), as summarized in Tables 1 and 2, respectively. Here, V is the volume of the dam, H f is the maximum residual thickness upon the completion of the dam failure, and T b is the time for the stage in front of the dam to rise up to the peak stage; T is the time from the onset to the end of the dam failure when the erosion of the residual thickness of the dam essentially finishes, which is estimated through the video of dam failure process, and T bg is the time for the initial breach to grow laterally to the full width of the channel (breach growth time).

Table 1 Summary of experiments for dams without initial breach
Full size table
Table 2 Summary of experiments for dams with initial breach
Full size table

A natural landslide dam may never be uniform across the channel. However, the “idealized” experimental cases for uniform dams are critical for revealing the fundamental mechanism of landslide dam failure and for calibrating and testing mathematical models (Cao et al. 2011). Likewise, the initial breach designed for the experiments (Fig. 2b) is in its geometry rather regular compared with natural landslide dams. Nevertheless, it facilitates easy implementation and is a reasonable surrogate for artificial spillway channels that could be dug out for alleviating downstream flooding under emergency conditions (such as that in the Tangjiashan Landslide dam following the 2008 major earthquake in Sichuan province, China).

The rate of natural landslide dam failure may vary significantly from case to case. Typically, the failure can last from several days (e.g., the June 1967 case in the Yalong River in China, see Li et al. 1986; Chen et al. 1992) to several years (Korup 2002; Costa and Schuster 1988). The quicker the landslide dam fails, the more severe the downstream flood will be. This study aims to focus on the hydraulics of the flood due to comparatively quick landslide dam failure, which is facilitated by relatively large discharges from the upstream (Tables 1, 2). Physical reasoning suggests that steep-faced dams can fail more easily and therefore generate more severe flood than mildly sloped dams, which is probably why previous experiments (e.g., Schmocker and Hager 2009) have focused on steep-faced dams. The present experiments are designed to be representative of relatively steep-faced dams, with height-to-length (measured along the channel) ratios varying between 1:9.5 and 1:5.5 (Tables 1, 2), which apparently falls within the range of 1:400 to 1:3.3 (Costa and Schuster 1988) of the height-to-length ratio of natural landslide dams.

4 Results and discussions

4.1 General description

Figure 4 shows a sequence of images of the failing dam and the flood flow at different times for F-Case 10. It is seen that the overtopping flow leads to quick failure of the dam, an observation consistent with that of Li et al. (1986), Chen et al. (1992) and Shang et al. (2003). Also, the dam erosion is almost uniform across the flume, which is probably dictated by that fact that the dam crest is initially level and the material is essentially homogeneous laterally. Generally, dam erosion initially occurs at the downstream surface of the dam. As the overtopping flow continuously erodes the dam, the dam top is rapidly lowered, and the downstream slope of the dam gradually becomes milder. It is also observed that the overtopping flow may induce active bed load sediment transport, which occurs essentially uniformly across the channel. For the dam without initial breach under the present experiments, overtopping erosion is indeed the primary factor inducing dam failure, and no substantial mass collapse could be observed. Basically, this observation is applicable to other experimental cases for dams without initial breach (Table 1).

Fig. 4
figure 4

Images of dam and flow at different times for F-Case 10, showing nearly uniform erosion across the channel. Overtopping erosion is the dominant factor for dam failure without initial breach, and no substantial mass collapse could be observed

Full size image

Figure 5 shows a sequence of images of the failing dam at different times for S-Case 6. The dam failure process substantially differs from its counterpart without initial breach (Fig. 4). Here, the dam failure begins within the initial breach, the flow erodes the downstream surface of the dam, and thus, an erosion gully is formed. Then, the gully is deepened gradually by continuous erosion. As the gully becomes progressively deepened and therefore its lateral slopes get steeper, mass lateral collapse of the gully occurs, which leads to gully widening. During this phase of the failure process, both erosion and mass collapse are important. In the later phase, as the initial breach has evolved adequately and the flow has fully overtopped the residual dam, mass collapse becomes less important and erosion overwhelms. For this particular case with a relatively large inflow discharge and also a sufficiently large initial breach (S-Case 6), no laterally full overtopping of the dam was observed before the initial breach is widened to the entire width of the channel. Yet, its occurrence is found in other cases (e.g., S-Case 3) with a comparatively smaller initial breach and differing discharge.

Fig. 5
figure 5

Images of dam and flow at different times for S-Case 6 with initial breach, showing gully deepening and widening by overtopping erosion and lateral mass collapse

Full size image

Figure 6 shows the measured stage hydrographs at four cross-sections and the degradation of the dam crest for F-Case 10. During the experiments, a ruler is set along the side glass wall of the flume to facilitate the recording of the level of the dam crest. This procedure provides estimated data for the degradation of the dam crest. While being approximate, this observation is rather effective, following Balmfroth et al. (2008). The degradation of the dam crest commences as the flow overtops the dam and ends shortly before the stage hydrographs at distinct cross-sections converge along the flume (Fig. 6).

Fig. 6
figure 6

Typical stage hydrographs and degrading dam crest (F-Case 10). The stage hydrographs can be divided into three distinct phases: the first sees stage rising upstream the dam (CS1 and CS5) prior to the onset of the dam failure, the second features stage falling and rising respectively up- and downstream (CS5 and CS12) the dam, and the third experiences gradual convergence of stages up- and downstream of the dam, following the completion of the dam failure

Full size image

The stage hydrographs illustrated in Fig. 6 may be divided into three distinct phases. The first phase sees continuous upstream stage rising before the onset of the dam failure, while the downstream stages (CS8 and CS12) remain at the initial static level. The second phase features continuous stage falling and rising respectively up- and downstream of the dam during the dam failure process; the third phase experiences gradual convergence of the upstream and downstream stages following the completion of the dam failure. During the first phase, the back and forth manner of the flood wave propagation is well exemplified in Fig. 6. Specifically, the stages rise asynchronously at differing cross-sections upstream of the dam, due to the continuous inflow from the inlet of the flume. The most upstream cross-section CS1 first sees stage rising and then CS5. Once the flood wave from the inlet reaches the dam, it is reflected by the dam and then propagates upstream, CS5 closest to the dam first experiences the reflected wave and thus stage rising, then CS1. In the second phase, the cross-section (CS5) closest to the dam first experiences the stage falling, followed by CS1. At a certain time, the stages at the cross-sections upstream of the dam start to converge. From the onset of the dam failure by overtopping, the downstream stages start to rise, and indeed, the stage at CS8 closest to the dam first rises and then at CS12 as the flood wave propagates toward the far downstream end of the channel. Also, the stages at distinct cross-sections downstream of the dam may reach peaks in a very short period of time, characterizing the highly transient nature of the flood. Following the completion of the dam failure process, in the third phase, the stages at different cross-sections, both up- and downstream of the dam, gradually converge and decline and finally reach a constant state.

In general, the stage hydrographs for all the experimental cases in Tables 1 and 2 qualitatively follow a pattern similar to those shown in Fig. 6. However, the time T b required for the upstream stage to reach the peak from the initial water level decreases with increasing inflow discharge, as illustrated in Table 1 for F-Cases 7, 8, and 9 and in Table 2 for S-Cases 4, 5, 6, and 7 (with roughly similar initial water levels). Indeed, this can be readily inferred from physical intuition. More importantly, the present experiments demonstrate that the time T from the onset to the end of the dam failure considerably decreases with increasing inflow discharge (F-Cases 7, 8, and 9, and S-Cases 4, 5, 6, and 7). This clearly characterizes the impact of the inflow discharge on the rate of the dam failure process. For a specific reservoir size, the higher the inflow discharge, the quicker the dam fails, and vice versa. From hydrodynamic perspectives, this observation can also be inferred, but has not been revealed in previous experimental studies (Coleman et al. 2002; Chinnarasri et al. 2003). Nevertheless, this understanding is attained under the present experimental conditions (flume), which, however, should not be construed as that other factors (e.g., capacity and irregular plan shape of reservoir) are negligible.

4.2 Influences of initial breach and surface slope

Figure 7 shows the measured stage hydrographs at selected cross-sections for F-Cases 9 and 15 (without initial breach) as well as for S-Cases 6 and 11 (with initial breach). It is shown that the stage upstream the dam without initial breach reaches a higher peak value, and the arrival of the flood in the downstream is appreciably delayed compared with the counterparts for cases with initial breach. Specifically, the arrival of the flood at CS8 (Fig. 7c) was delayed by about 130 s and 70 s when comparing F-Case 9 and S-Case 6, and F-Case 15 and S-Case 11. This is exclusively ascribed to the initial breach, which features a base lower than the crest of the dam. With initial breach, it takes less time for the stage upstream the dam to reach the base level of the initial breach than the dam crest, the failure onsets earlier by the overtopping flow, and consequently, in the upstream, the peak stage is reduced, and in the downstream, the flood arrives sooner than the case without initial breach. Figure 7 also shows that when there is initial breach, the peak stage in the downstream is detectably lower than its counterpart for cases without initial breach, which in principle means mitigated flooding. While the difference in the downstream peak stages for cases without and with initial breach is limited for these particular cases, it can be considerably pronounced at a lower inflow discharge (Fig. 8 for F-Cases 7 and 12 against S-Cases 4 and 8). Further, the impact of initial breach over the stage hydrographs is appreciably attenuated by decreased slopes of the initial dam surface. This can be found in Fig. 7, as the difference between F-Case 15 and S-Case 11 is reduced as compared against that between F-Case 9 and S-Case 6. In general, the differences between the stage hydrographs for dams with and without initial breach (Fig. 7) are qualitatively applicable to other cases (Tables 1, 2).

Fig. 7
figure 7

Stage hydrographs for F-Case 9 and S-Case 6 (S 1 = 1:2, S 2 = 1:3) as well as for F-Case 15 and S-Case 11 (S 1 = 1:4, S 2 = 1:5), showing effects of the slopes of initial dam surface as differentiated by initial breach

Full size image
Fig. 8
figure 8

Stage hydrographs for F-Case 7 and S-Case 4 with non-cohesive sand as well as for F-Case 12 and S-Case 8 with sand–clay mixture, showing that under a relatively low inflow discharge, the effect of clay is substantial when there is no initial breach (F-Cases 7 and 12) and in contrast marginal as overwhelmed by initial breach (S-Cases 4 and 8)

Full size image

The slopes (S 1 and S 2, Fig. 1(a)) of the initial dam surface may act to modify the failure process. Indeed, this is revealed by the present experiments. When there is no initial breach (F-Cases 9 and 15), reduced slopes of the initial dam surface seem to exhibit stabilizing effects. This is reflected by not only the larger value of the maximum residual thickness H f (Table 1) for F-Case 15 (than F-Case 9), but the stages. While marginal from the stage hydrographs downstream the dam (CS8 and CS12), the slope effects are appreciable from the stages in the upstream (CS1 and CS5), see Fig. 7. The rate of the declining of the upstream stage for F-Case 15 is lower than F-Case 9, and especially, the final stage of F-Case 15 is a bit higher than F-Case 9. The later observation is not surprising at all from the point of view of open channel hydraulics. After the final phase of the dam failure process, the flow is nearly steady because the inflow discharge is constant, and the flow upstream the dam is entirely controlled by the inflow discharge and the residual thickness. The higher residual thickness for F-Case 15 raises the water level in comparison with F-Case 9. In contrast, downstream the (residual) dam, the discharge is the same as from the upstream, therefore the final stages almost converge for F-Cases 15 and 9 (CS8 and CS12, Fig. 7).

When there is initial breach (S-Cases 6 and 11), the slope effects are qualitatively similar to those under the cases without initial breach (F-Cases 9 and 15), but quantitatively enhanced to a certain extent, as shown in Fig. 7. In particular, because of the stabilizing effects of the milder slopes of S-Case 11, the development of the initial breach is largely hampered (than S-Case 6), the discharge through the dam during this phase is much lower than the inflow discharge, and accordingly, the upstream stage continues to rise over the base level of initial breach after the onset of the failure processes. This is why the peak stage of S-Case 11 upstream the dam (CS1 and CS5) is much higher than its counterpart for S-Case 6 (Fig. 7). This observation appears to characterize the highly complicated nature of the impacts of dam geometry on the failure process.

4.3 Influence of dam composition

When the dam is composed of sand–clay mixture or sand–gravel mixture (Tables 1, 2), the failure process and the stage hydrographs of the flood are qualitatively similar to those for dams of purely non-cohesive sand (Figs. 4, 5, 6). Briefly, under the present experiments, overtopping erosion is the primary factor inducing the dam failure, and lateral mass collapse is also considerable where there is initial breach. Quantitatively, the clay content may appreciably depress the rate of the dam failure process and accordingly increase the total time of the dam failure process, if there is no initial breach. As illustrated in Table 1, the time T from the onset to the completion of the dam failure increases as clay is added to the dam (F-Case 14 versus F-Case 9, F-Case 13 versus F-Case 8, and F-Case 12 versus F-Case 7, Table 1), while other conditions remain unchanged. This observation of the clay effect is more obvious when the inflow discharge is lower. In this connection, the time T evidently decreases with increasing inflow discharge, whether there is clay or not (Tables 1, 2).

Figure 8 shows a comparison between the measured stage hydrographs for F-Cases 7 and 12 as well as for S-Cases 4 and 8. These four cases feature the same inflow discharge from the upstream, 0.006 m3/s, and approximately the same initial water depths, but different dam composition (with or without cohesive clay) and initial dimension of breach (with or without initial breach), see Tables 1 and 2. Figure 8 shows that there is substantial difference between the stage hydrographs for F-Cases 7 and 12 without initial breach, exemplifying the role of clay in modifying the dam failure process and the flood flow. Particularly, the peak stages upstream of the dam for F-Case 7 are considerably lower than those for F-Case 12 (~3.5 cm at CS1 and CS5), and the flood wave for F-Case 7 arrives considerably earlier (~260 s at CS8 and CS12) in the downstream than its counterpart for F-Case 12. Physically, seepage through the dam is inevitable, especially when there is no cohesive clay in the dam (purely non-cohesive sand). While the amount of water loss arising from the seepage may be rather limited compared with the bulk volume of the water in the reservoir, the seepage may lead to substantial dam subsidence prior to the onset of the dam failure by overtopping. This effect can be augmented for those experimental cases with relatively small discharges and high dams (e.g., F-Case 7), because of the longer time needed for the water level in the reservoir to reach the dam crest to initiate overtopping erosion. As the dam is substantially subsided, the onset of the dam failure by overtopping occurs earlier and at an appreciably lower upstream stage, and therefore, the flood wave reaches a downstream cross-section much earlier (Fig. 8). However, when there is initial breach in the dam, the overall difference is largely reduced or simply marginal between the stage hydrographs in relation to dams of purely non-cohesive sand and sand–clay mixture (S-Case 4 versus S-Case 8). Consequently, the impacts of cohesive clay over the dam failure process and the flood are readily overwhelmed by the initial breach.

The impacts of cohesive clay on the dam failure process and the flood can be modified by the inflow discharge from the upstream. Figure 9 shows a comparison between the measured stage hydrographs for F-Cases 9 and 14 as well as for S-Cases 6 and 10. These four cases have the same inflow discharge from the upstream (0.025 m3/s), much higher than those shown in Fig. 8 (0.006 m3/s) and approximately the same initial water depths, but different dam composition (with or without cohesive clay) and initial dimension of breach (with or without initial breach). As shown in Fig. 9, the stage hydrographs between F-Cases 9 and 14, as well as between S-Cases 6 and 10 exhibit little difference; thus, the effect of clay is overwhelmed by the higher inflow discharge.

Fig. 9
figure 9

Stage hydrographs for F-Case 9 and S-Case 6 with non-cohesive sand as well as for F-Case 14 and S-Case 10 with sand–clay mixture, showing that under a comparatively large inflow discharge, the effect of clay is negligible irrespective of whether there is initial breach (S-Cases 6 and 10) or not (F-Cases 9 and 14), as overwhelmed by the inflow

Full size image

Figure 8 also indicates that the initial breach may have substantial impacts on the flood. The downstream peak stages for S-Cases 4 and 8 are much lower than those of F-Cases 7 and 12, respectively (by 2.7 and 3.4 cm at CS12, i.e., by 17.31 and 20.41%). And yet, when the inflow discharge increases, the difference between the downstream peak stages due to initial breach diminishes. As shown in Fig. 9, the downstream peak stages for S-Cases 6 and 10 are lower than those of F-Cases 9 and 14, respectively, by 1.3 and 2.0 cm at CS12 (i.e., by 7.32 and 10.80%). Comparatively, a higher inflow discharge leads to higher peak stages in the downstream, while other conditions remain unchanged. As shown in Figs. 8 and 9, the peak stage at CS12 for F-Case 9 is higher than its counterpart for F-Case 7 by about 2.3 cm (i.e., by 14.74%).

While the difference in the downstream peak stages due to clay, initial breach and inflow discharge is of the order of magnitude of centimeters under the laboratory scales, it may point to considerably varied flooding in the field prototype scale and merits consideration in flood defense design and risk management. The above observations certainly characterize the prime role of the inflow discharge in dictating the dam failure process and the flood, which can overwhelm the impacts of cohesive clay and initial breach.

The impacts of gravel in the dam on the failure process and the flood are substantially different from those of cohesive clay (Figs. 8, 9). Figure 10 shows a comparison between the measured stage hydrographs for F-Case 15 and S-Case 11 with non-cohesive sand as well as for F-Case 16 and S-Case 12 with sand–gravel mixture. Irrespective of whether there is initial breach or not, the gravel exhibits stabilizing effects and therefore affects the flood. The increased maximum residual thickness H f due to gravel (F-Case 16 versus F-Case 15, S-Case 12 versus S-Case 11) is certainly a manifestation of this effect (Tables 1, 2). Though the difference between the downstream (CS8 and CS12) stage hydrographs is only marginally appreciable for F-Cases 16 and 15 or S-Cases 12 and 11, it is considerable in the upstream (CS1 and CS5), as shown in Fig. 10. Apparently, the rate of the declining of the upstream stage for F-Case 16 and S-Case 12 is respectively lower than F-Case 15 and S-Case 11. Also, the final stage of F-Case 16 is a bit higher than F-Case 15 in the upstream (CS1 and CS5), so is that for S-Case 12 compared with S-Case 11. Not surprisingly, the later observation is actually dictated by a mechanism similar to what leads to the higher final stage in response to reduced slopes of the initial dam surfaces (Fig. 7). More specifically, it is the higher residual thickness of F-Case 16 and S-Case 12 that raises the water level in the upstream, as compared against F-Case 15 and S-Case 11.

Fig. 10
figure 10

Stage hydrographs for F-Case 15 and S-Case 11 with non-cohesive sand as well as for F-Case 16 and S-Case 12 with sand–gravel mixture, showing stabilizing effects of gravel, irrespective of whether there is initial breach or not

Full size image

Practically, it is interesting to see how quickly the initial breach is widened in response to different inflow discharges. Figure 11 shows the breach width growth in time for S-Cases 4, 5, 6, and 7. All the four cases feature the same dam composition (non-cohesive sand) and initial dam geometry (Table 2). As can be derived from physical reasoning, a high-inflow discharge leads to earlier onset of the breach widening and quicker breach growth and vice versa (Fig. 11). More specifically, the time T bg required for the initial breach to grow laterally to the full width of the channel decreases with increasing inflow discharge (Table 2), as approximately estimated using the video recordings of the dam failure process. Under the present experiments, the impact of dam composition (with clay or gravel) on breach growth is appreciable but minor than that of the inflow discharge, see S-Cases 4, 5, and 6 as compared against S-Cases 8, 9, and 10, respectively, and S-Case 12 in comparison with S-Case 11 (Table 2). Breach widening is further addressed along with computational modeling in the companion paper (Cao et al. 2011).

Fig. 11
figure 11

Breach width growth in time for S-Cases 4, 5, 6, and 7 in relation to differing inflow discharges (Table 2). It is shown that a higher inflow discharge leads to earlier onset of breach widening and quicker breach growth

Full size image

5 Conclusion

An experimental investigation is presented of landslide dam failure and the flood flow. Under the present experiments, overtopping erosion is the primary mechanism of landslide dam failure, and lateral mass collapse is also considerable where there is initial breach. This is significant as natural landslide dams are laterally uneven and in cases artificial channels are generated for alleviating the downstream flooding. The inflow discharge dictates the dam failure process and the flood, and the dimension of initial breach is also essential. Milder slopes of the initial dam surface and gravel content in the dam can considerably stabilize the dam, depress the dam failure, and thus affect the flood. Cohesive clay may act to mitigate the seepage through the dam and its subsidence and also appreciably modify the timing of the onset and the rate of the dam failure process. Yet, the impacts of cohesive clay can be readily overwhelmed by a high inflow discharge and initial breach. The observations highlight the vital importance of (a) securing reliable information of the inflow discharge and (b) generating an initial breach of sufficient dimensions for effective management of the downstream flood. The present work provides new observational laboratory data on landslide dam failure and the flood, useful for calibrating and testing mathematical models of the processes.

The mechanism and data of landslide dam failure and the flood attained from this study are inevitably constrained by the spatial and time scales that can be realistically accommodated in the experimental facility. The potential impact of anisotropy of natural dam composition is not yet explored. The impacts of inflow discharge, initial breach, dam geometry, and composition on the flood set within the laboratory scales can be expected to quantitatively amplify in the field prototype scale. In principle, the similitude theory for physical hydraulic modeling (Allen 1952) can be deployed to project the laboratory stage hydrographs to prototype scales, though it needs to exercise due caution, especially when the understanding of sediment transport and mass collapse remains far from complete and the quantification bears uncertainty. This is addressed in the companion paper, as related to mathematical modeling (Cao et al. 2011) that is not subject to scale constraints. While being difficult and even risky to accomplish, field observations of landslide dam failure and the flood will be of significance, especially those that can provide more systematic data than previous studies (Hoeg et al. 2004) for the dam failure and the flood stages in a sufficiently long reach.