Detachment and Initial Motion

The story begins at the instant that a rock mass (which is essentially intact) becomes detached from its parent rock edifice due to completion of a failure surface by crack propagation and coalescence. At this instant the rock mass begins to accelerate under gravity by moving along its failure surface—generally by sliding, but for smaller events toppling or falling may play a role. The initial motion seems likely to be as an initially more-or-less intact mass that slowly begins to disaggregate at joint scale due to sliding over a non-planar, non-circular, and irregular surface. This stage of motion was well captured by the well-known set of still images of the 1980 Mount St. Helens failure taken by Gary Rosenquist (e.g., Lipman & Mullineaux, 1981; Mount St. Helens Erupting); the slow breakup of the huge mass (prior to release of internal volcanic pressure and the resultant explosion) is clear. At this stage the lack of internal fragmentation is evident from the lack of dust ejected from the moving mass.

Collapse

As the sliding mass leaves its initial location and moves down the steep upper part of the edifice, with increasing velocity and displacement, progressively smaller-scale collapse of the mass ensues in which the rock breaks into largely joint-controlled fragments. This transforms the initial, partly intact rock mass into a granular mass whose lower resistance to deformation facilitates translation and which can thus accelerate faster. During this stage fragmentation will begin to occur; this term refers to the breakage of intact rock by creation and extension of new cracks as moving grains stress each other and the underlying surface. As a result of fragmentation very small rock fragments are generated that begin to appear as a dust cloud forming on the translating granular mass. The Piz Cengalo video shows the transition from sliding to flowing and the start of fragmentation.

Flow

When the collapsing and partly fragmenting granular mass reaches the lower and less steep parts of its parent edifice, its constituent grains will increasingly flow in more-or-less continuous contact with each other as a dense granular flow (e.g., MiDi, 2004) with a well-defined upper surface. During this phase the intergranular contact stresses are a maximum at the base of the flow (here assumed to be rigid for simplicity), which is also the location of the greatest shear, and fragmentation becomes intense here as grains forced to shear past each other cannot dilate upwards to do so, resulting in grain breakage. As the flow accelerates (and thins due to longitudinal extension), shear and fragmentation extend upwards until all but the upper few meters are fragmenting intensely. In the upper few meters, the overburden pressure is low enough that grains can shear past each other without breaking, forming a less-fragmented layer known as the carapace. In the flow phase the presence of intense fragmentation is evident from the voluminous dust cloud emitted from the flow.

Deceleration and Halting

Eventually (in this idealized scenario) the gradient of the runout path becomes low enough that the front of the granular flow (the thinnest part) begins to decelerate; this causes the following material to slow, sometimes creating lateral compression ridges. As the distal edge comes to halt, the following material also does so. The dust cloud may persist for hours to days, spreading widely.

Precursory Processes

A rock avalanche initiates when a large mass of rock detaches from its position as part of a mountain edifice. This mass of rock has previously been gradually raised to its final position, usually near the top of the mountain, by tectonic uplift. Part of this process involves the development and extension of a network of cracks within the rock of the mountain, many of which eventually link up to form a continuous crack system that effectively separates the future avalanche mass from the rest of the mountain. Often, the rock mass that fails and causes a rock avalanche is less damaged than the surrounding rock, which has eroded more rapidly in smaller events to leave the large protruding mass of sounder rock able to fail en masse. Once separated from the mountain by a continuous failure surface, the mass may be unstable and start moving, or it may require some additional force, such as reduced friction due to rainfall, or a seismic force due to an earthquake, to set it in motion.

A further precursory process that is often invoked to explain rock avalanche occurrence is glacial debuttressing. This occurs during deglaciation of a mountain region, in which downwasting of ice masses in valleys leave high rock walls, previously flanked and “buttressed” by ice, exposed to both weathering and failure processes. In some ways the role of debuttressing has been overstated (e.g., McColl, Davies, & McSaveney, 2010), because ice is a fluid of very high viscosity and low tensile strength and as such cannot resist the stresses imposed by a failed slope; the presence of valley ice, however, will slow the motion of a failed mass, and the rapid downwasting of ice may conversely allow acceleration of motion to form a rock avalanche, particularly if the failure surface becomes exposed to air rather than daylighting beneath the ice (McColl, Davies, & McSavenery, 2012). Ice covering mountain topography to great depth would reduce the coseismic deformation of the mountain because under high-frequency shaking ice acts as a strong and rigid solid; downwasting of the ice surface would then increase the ability of the mountain to respond to seismic shaking (McColl & Davies, 2013) and thus to develop deep-seated failure surfaces triggering rock avalanches.

Triggers

Spontaneous Failure

The rock comprising the upper part of a mountain is always under high gravitational stress due to its topography and thus tends to move downwards; for example, long linear antiscarps (or Sackungen) on mountainsides in a range of environments reflect the gradual gravitational settlement of the mountain by deformation on suitably aligned joint sets (e.g., McColl, 2012; Figure 12). This demonstrates that all mountain rock masses experience ongoing crack development and extension, even though the stresses in the rock are much lower than the failure strength of the rock. Thus the rock is continuously getting weaker as cracks extend and coalesce—this process is called “stress corrosion” or “static fatigue” (e.g., Anderson & Grew, 1977). In the absence of any external trigger, stress corrosion can cause a mountainside to collapse entirely spontaneously—there have been six landslides in the Southern Alps of New Zealand since the Aoraki/Mt. Cook rock avalanche of 1991, all about 1 to 10 million cubic meters in volume, none of which were associated with either rainfall or seismicity, so they may have been the result of stress corrosion. Of course, there may well have been triggers below the limit of detection such as freeze-thaw–generated stresses, but such apparently spontaneous failures are by no means rare. Aseismic rock avalanches tend not to have deep-seated headscarps, rather the failing masses tend to be extensive and relatively shallow (e.g., Figure 13; the other recent aseismic rock avalanches in the Southern Alps also have source areas similar in morphology to the adjacent hillslopes).

Open in new tab

Figure 12. Gravitational antiscarps on Beinn Fhada, Kintail Scotland.

Source: Verne Pere.

Open in new tab

Figure 13. The light-colored area near the top of Mt. Adams (outlined by black dashed line) is the source of the 10⁷ m³ aseismic rock avalanche that formed the Poerua landslide dam in 1999. Note the lack of a topographically defined source compared with Figures 2, 3, and 7.

Earthquakes

By contrast with rainfall, rock avalanches are frequently caused by earthquakes. Nevertheless, even in seismically active regions it cannot be assumed that all large landslides are coseismic. The six large aseismic landslides in the Southern Alps of New Zealand between 1991 and 2013 suggest that caution is required in such assumptions. Indeed, were such a cluster of large landslides to be found in the paleo record, the interpretation would almost certainly be that a seismic trigger was involved—but it was not.

There are reasons for expecting coseismic landslides to be larger in volume than their aseismic counterparts. When an earthquake shakes a mountain, the shaking frequency that contains most of the seismic energy is in the range of 1 to 10 Hz; this corresponds to the natural frequency of oscillation of mountains with hundreds to thousands of meters of relief above their surrounding valleys (Buech, Davies, & Pettinga, 2010). Hence we should expect seismic excitation to cause a mountain to oscillate with increasing amplitude at its natural frequency (Bazgard, Buech, & Davies, 2009), and this motion is likely to cause potential failure surfaces to develop at considerable depth (maybe hundreds of meters) within the mountain. If failure occurs on these surfaces it will involve more volume—for a given failure area—than a shallow near-surface failure caused by high pore-water pressures. As an example, Barth (2014) reported a 0.7 km³ rock avalanche in the Cascade valley of South Westland, New Zealand, associated with a very large subsidence in the adjacent hillslope (Figure 14). It seems likely that prior to failure the rock avalanche mass had a similar morphology and that over many seismic events (the plate-boundary Alpine fault runs at the base of the slope and generates M8 earthquakes every few hundred years) the mass had slumped progressively as a failure surface propagated deeply below the slope, until eventually the failure surface daylighted and caused failure. Chigira, Tsou, Matsushi, Hiraishi, and Matsuzawa (2013) investigated a set of large slope failures that occurred during Typhoon Talas in Japan, finding that many of these had pre-failure morphology similar to that displayed by the Cascade slope, so it is possible that these rainfall-triggered failures had been preconditioned by earthquakes. In this case, the rainfall-triggered rock avalanche is likely to have a volume similar to that of a coseismic rock avalanche from the same source.

Open in new tab

Figure 14. Cascade rock avalanche, Westland, New Zealand. Source area = dashed line; deposit = dotted line; Alpine fault = chain dashed line. Note the prominent antiscarp (arrowed) to the right of the headscarp and the bulging hillslope beneath it. The failure surface is up to about 250 m below the ground surface.

Source: Modified Google Earth image.

An interesting characteristic of some coseismic rock avalanches is that they involve the summit or ridge of the parent mountain—that is, the deep-seated failure surface involves the top of the mountain and daylights on the opposite flank. This often results in the source area being marked by a distinct dip in a ridge-line (Figure 15); since aseismic landslides seem not to involve ridge-crests, this feature may be a useful indicator of seismic activity when a deposit lies immediately below.

Open in new tab

Figure 15. This rock avalanche deposit near Mesopotamia, South Island, New Zealand, has a prominent dip in the ridgeline immediately above it (dashed line), indicating that the failure surface daylighted on the far side of the ridge. Note the apparently short runout, despite a volume of about 60 × 10⁶ m³—perhaps due to later river trimming of the deposit.

Source: Google Earth image.

Low Apparent Friction

When a solid block slides down a concave slope under gravity (Figure 16), its motion is limited by the coefficient of friction (µ‎) between the slope and the block; the block will come to rest at such a location that the line joining the initial and final positions of its center of mass is inclined below the horizontal at an angle of φ‎, where

In this case the Fahrböschung or travel angle ψ‎ (defined as the angle to the horizontal of a line joining the highest point of the headscarp to the distal edge of the deposit) is almost exactly parallel to the line joining the centers of mass, because the block has not changed shape while sliding. However if, like a rock avalanche, the initial block spreads longitudinally (and laterally—but we are only thinking 2D at the moment) during motion (Figure 17), then the situation changes markedly; while the displacement of the center of mass of the material may be similar (assuming µ‎ is the same in both cases), the distal end of the deposit has now traveled much farther. Here the travel angle ψ‎ is much lower than φ‎. This effect has often been interpreted as reflecting low basal friction, whereas in fact it simply results from internal deformation of the debris as it travels. Thus a low travel angle does not necessarily indicate low friction; indeed, Davies, McSaveney, and Hodgson (1999) measured ψ‎ as low as 12° in a laboratory fall of one liter of fine dry sand whose basal and internal friction angles were both ~27°.

Open in new tab

Figure 16. Frictional sliding of a rigid block. φ‎ is the friction angle; ψ‎ is the Fahrböschung.

Open in new tab

Figure 17. Frictional sliding plus longitudinal spreading of a deforming landslide.

Many large rock avalanche deposits, however, have φ‎ (as indicated by center-of-mass displacement) considerably lower than the normal friction angles of granular materials and of rock-on-rock sliding and correspondingly lower ψ‎ values (Figure 18; acknowledging, however, that φ‎ is difficult to infer accurately in three-dimensional deposits whose basal geometry is poorly constrained). In these cases there is an anomalously low frictional resistance to the travel of the debris. The primary puzzle surrounding rock avalanches is why friction is low in large rock avalanches.

Open in new tab

Figure 18. Low-friction sliding plus spreading of a rock avalanche; note lower values of φ‎ and ψ‎.

Suggested Mechanisms

Proposed explanations for the long runout of rock avalanches to date include: air-layer lubrication (Shreve, 1968b); air-extrusion fluidization (Kent, 1966); fluidization by a dense interstitial dust dispersion (Hsü, 1978); generation of steam by frictional heating of pore water (Goguel, 1978); gaseous pore pressures (Habib, 1975); basal sliding on dissociated or melted rock (De Blasio & Elverhoi, 2008; Erismann, 1979; Legros, 2002; Masch, Wenk, & Preuss, 1985); development of a low-density high-vibration basal layer (Campbell, 1989; Campbell, Cleary, & Hopkins, 1995; Johnson, Campbell, & Melosh, 2016); shear of a wet basal zone (Voight & Sousa, 1994); basal pressure wave propagation (Kobayashi, 1997); lubrication by undrained loading of saturated substrates (Abele, 1974; Legros, 2002; Sassa, 1988); mass changes (Hungr & Evans 2004; Van Gassen & Cruden, 1989); mechanical fluidization (Davies, 1982; McSaveney, 1978); acoustic fluidization (Collins & Melosh, 2003; Melosh, 1979); random fragment kinetic energy (Preuth, Bartelt, Korup, & McArdel, 2010); seismic energy fluidization (Hazlett, Buesch, Anderson, Elan, & Scandone, 1991); oscillation of quasi-rigid plugs (Cagnoli & Quareni, 2009); and dynamic rock fragmentation (Davies & McSaveney, 2002, 2009; Pollet & Schneider, 2004).

Davies and McSaveney (2012) examined the above mechanisms on the assumption that there is a single mechanism that explains long runout in all the circumstances and environments in which it occurs. Davies and McSaveney (2012) concluded—on this basis—that the only viable mechanism that was applicable in the context of all rock avalanche characteristics and environments is that of dynamic rock fragmentation.

Space precludes a complete discussion of all these mechanisms here. Instead, five mechanisms are outlined and discussed, which are either representative of a broader class or are of particular historical significance.

First, air-layer lubrication, proposed by Shreve (1968a), was an early and ambitious attempt to explain some of the characteristics of the prehistoric Blackhawk landslide (Shreve, 1968b) and the 1964 Sherman Glacier supraglacial rock avalanche (McSaveney, 1978); both of these deposits were unusually thin relative to their area compared to other deposits (the average depth of the 8 km² Sherman Glacier deposit was 1.65 m; McSaveney, 1978), and both moved over substantial cliffs early in their motion (as, indeed, did the Elm event; Heim, 1882). These factors led Shreve to suggest that the moving debris, having been “air-launched” over the cliff, trapped a layer of compressed air beneath it when it fell to the substrate, which effectively separated the debris from the substrate, allowing it to move as a non-shearing sheet with very low basal friction—much like a hovercraft. Other features such as lateral and terminal levees could also be explained by this mechanism. The idea enjoyed considerable attention but has fallen from favor as long-runout deposits have been discovered on the Moon, where the mechanism is not feasible in a hard vacuum, and as concern has increased about the required impermeability of the moving granular mass to air at high pressure. Nevertheless, no formal refutation of the mechanism has been published to date.

Another very popular class of mechanisms for explaining the long runout of rock avalanches focuses on the nature of the substrate over which the runout occurs. Evidently the direct and shear forces applied suddenly to a weak substrate as a result of the passage of the front of a large rock avalanche tend to cause the substrate to compress, and if the substrate comprises saturated material then its pore-water pressure can become very high and its resistance to shear correspondingly low due to undrained loading (e.g., Abele, 1974; Sassa, 1988; Legros, 2002). Since many rock avalanches fall into mountain valleys, and these are often are floored with saturated alluvial fills, this mechanism has obvious potential relevance. A complication with this concept, however, is the necessary erosion of the substrate material that has been weakened by the sudden loading as the rock avalanche front affects it; this means that for the low friction resulting from liquefaction to persist while the rest of the rock avalanche passes the same location, liquefaction (and erosion) must continue to occur in the substrate material underlying that which has been eroded and incorporated into the avalanche. Further, concentration of shear within the substrate means that the rock avalanche debris is transported passively with negligible shear, which raises the question of how the intense fragmentation ubiquitous in rock avalanches is generated.

Van Gassen and Cruden (1989) assumed that the mass of debris in motion reduces during runout, as debris is deposited on the runout path. If the original momentum of this deposited material is transferred to the still moving material, as assumed by Van Gassen and Cruden (1989), then the momentum per unit mass of the moving debris increases, thus so does the mass velocity, and the deceleration of the moving debris due to friction is reduced. This mechanism is entirely internal to the mass in motion and can operate irrespective of environment. However, closer examination reveals the key assumption that, in order for the momentum of the deposited material to be transferred to the still moving debris, the depositing material must be ejected rearwards from the moving mass at a velocity equal to that of the moving mass. That is, it must act like rocket fuel and be forced backwards relative to the moving mass in order to transfer its momentum to the moving debris. No known mechanism allows this to occur. By contrast, the frequent observation that rock avalanche debris in motion incorporates material from the substrate requires that the moving material transfers part of its momentum to originally stationary substrate material, hence reducing the momentum per unit mass and the mean velocity of the mass in motion.

Melosh (1979) introduced the concept of acoustic fluidization from chemical engineering to geological sciences, in particular to impact cratering, earthquakes, and long-runout rock avalanches (e.g., Collins & Melosh, 2003). The essence of this concept is that unbreaking grains in a shearing granular flow are forced to move past each other and past an assumed rigid substrate, and this causes them to oscillate perpendicular to the flow direction; integrated over the whole moving mass this results in a general vibrational stress field superimposed on the mean flow. This vibration causes the interparticle contact stresses to be time-variant, and thus also the interparticle frictional resistance; so there are times when particles can shear past each other, and past the substrate, very easily, while when interparticle stresses are temporarily high, shearing is inhibited. This concept is similar to the mechanical fluidization suggested by McSaveney (1978) and Davies (1982); to the occurrence of a high-dilatation low-density and low-friction basal layer suggested by Campbell (1989); and to the concept of random fragment kinetic energy (Preuth et al., 2010). Acoustic fluidization has been examined in detail in the context of earthquakes by Sornette and Sornette (2000), with the outcome that shear-induced vibration is found to be insufficiently energetic to reduce the friction to the required extent, and the same conclusion seems likely to apply also to rock avalanches.

Davies, McSaveney, and Hodgson (1999) and Davies and McSaveney (2002, 2009) proposed and developed the concept that dynamic rock fragmentation during rock avalanche runout can explain the excess travel distance in principle, and Davies, McSaveney, and Kelfoun (2010) demonstrated that the concept can quantitatively explain the well-constrained deposit geometry of the Socompa volcanic debris avalanche. Dynamic rock fragmentation is the breakage of intact rock clasts in the interior of a rock avalanche due to shearing under high direct stress—instead of moving laterally and vertically to allow shear without breakage, the clasts are forced to shatter. This shattering occurs explosively (as in a rock-burst), and the simultaneous shattering of many clasts of all sizes throughout the debris generates a pervasive high-frequency high-stress elastic body-wave field—much like shear-induced vibration in acoustic fluidization, but the field, generated by fragmentation is about 1,000 times more intense than that generated by shearing alone (Davies & McSaveney, 2009). This field acts like a high pore pressure to reduce the interparticle stresses and hence intergranular friction. Criticisms of this concept are based on incomplete debris particle-size distributions (e.g., Crosta, Frattini, & Fusi, 2007; De Blasio & Crosta, 2015; Locat, Couture, Leroueil, Locat, & Jaboyedoff, 2006) that extend only to about 1 micrometer (10⁻⁶ m), whereas most of the particles in rock avalanches debris are at least an order of magnitude smaller than this (McSaveney & Davies, 2007).

Idealized Dry Rock Avalanche, Inerodible Substrate

We consider a rock avalanche in the middle of its runout; it is entirely granular at this stage, all of the original rock mass having collapsed on pre-existing joints, and for simplicity we assume that the flow is uniform (constant along-flow) and steady (time-invariant). We assume that the flow is driven by a base-parallel force on each grain, and the flow develops to a steady state in which the along-flow force balances the basal friction drag. The basic process is that of a shearing granular flow moving across a rigid uneven substrate. Numerical models of such flows are common, usually with unbreakable circular or spherical grains of identical diameter that interact according to predetermined rules, and with a substrate geometry similarly simply defined; the models commonly show the presence of a basal region of rapidly shearing highly dilated grains in which intergranular friction is lower than average and most of the shear takes place, overlain by more slowly shearing material (e.g., Campbell et al., 1995; Johnson et al., 2016). Thus, while this basic model is the simplest possible, and its simplifications are considerable, the behavior that emerges is unexpectedly complicated and depends significantly on factors such as grain interactions and ratio of flow depth to grain diameter. Nevertheless, the model can be shown to result in long runout (e.g., Cleary & Campbell, 1993).

Non-Uniform Grains

If the grain-size distribution of the granular flow is now assumed to be non-uniform, then vertical segregation of different-sized grains occurs; due both to kinetic sieving (whereby small grains can fall downwards between larger grains) and to dynamic segregation (whereby larger grains experience larger upward forces and thus move upwards; e.g., Bagnold, 1954), larger grains tend to accumulate in the upper parts of the flow and smaller grains in the lower. This causes the along-flow force balance to alter, and different flow characteristics (e.g., velocity distribution) result.

Erodible Substrate

If the substrate is erodible, the situation alters significantly. First, erosion of the substrate means that the condition of uniform flow no longer applies because the mass in motion increases along-flow and, because erosion depth increases with time at a given location, flow is also time-varying. Next, the grain-size distribution of the substrate relative to that of the rock avalanche debris affects the composition of the moving mass even if the substrate is dry and granular and may affect it even more fundamentally if it contains water or is a plastic solid. Hence the substrate composition has a fundamental influence on the flow behavior. Field data emphasize this; for example, rock avalanche deposits emplaced onto saturated substrates often show longitudinally aligned hummocks (Dufresne & Davies, 2009). An extreme example is that of a snow/ice substrate; rock avalanches emplaced on such materials generally behave quite differently from those emplaced on earth materials, having much thinner, wider and longer deposits which are noticeably grooved longitudinally. An interesting video of the 2013 Mt. Dixon supraglacial rock avalanche, New Zealand (Hancox & Thompson, 2013), shows the initial fall from a high ridge generating plentiful dust, after which the dust generation ceases and the front of the moving mass, with much entrained snow and ice, is seen to be sliding as a coherent mass and slowing to a halt. Another video shows the whole deposit extremely well, which is interesting to compare with imagery of the 1964 Sherman Glacier rock avalanche (McSaveney, 1978)—both show extremely well-developed longitudinal ridging.

Although the basal shear stresses of a rock avalanche would be expected to be very high, this is not necessarily the case. Figure 19 shows evidence that a very large volcanic debris avalanche passed over a peat layer formed on sand but did not completely remove it, even though the event traveled several kilometers further, so even soft substrates may be not eroded. This in fact corresponds well in principle to the occurrence of low basal friction—which itself implies low basal shear stress—allowing long runout.

Open in new tab

Figure 19. Basal contact of a large volcanic debris avalanche from Mt. Taranaki, exposed at Bell Beach, New Plymouth, New Zealand. The grey material immediately beneath the clast-studded avalanche debris is peat, still in place above reddish beach sand. The avalanche ran out about 5 km beyond this location. This contact is traceable for about a kilometer along the coast. Width of view about 0.5 m.

Source: Photo by Anja Dufresne.

Interstitial Fluid

While the copious dust clouds emitted by most rock avalanches suggest that the phenomenon is essentially dry, rock avalanches in some environments are significantly affected by interstitial fluids. Large submarine deposits show many of the characteristics of their terrestrial equivalents, such as prominent hummocks, and evidently are mostly if not totally saturated throughout their travel; and volcanic debris avalanches, which originate in edifices containing substantial quantities of water, have similar deposits to dry rock avalanches, but with differences in hummock size relative to depth and of runout relative to volume (Davies & McSaveney, 2009). The presence of water in a moving mass of rock avalanche debris will inhibit the relative motion of individual grains due to the viscosity of the water, and if the water forms a slurry with the fine fraction of the debris, the viscosity may be much higher. In addition, pore water pressure reduces the intergranular contact pressure and may affect intergranular friction in the interior of the flow; at the front and edges, especially with polydisperse grains, levees tend to form because the debris there may be unsaturated. Nevertheless, in a dry rock avalanche interstitial fluid effects are likely to be due mainly to air, which is of low density and viscosity so that its effects are minor.

Grain Fragmentation

The final complication considers the grain fragmentation that occurs ubiquitously in rock avalanches of all types and in all terrestrial environments and is evidenced by the large dust clouds that accompany subaerial rock avalanches. This means that the grain-size distribution alters progressively during runout. Again this complicates the dynamics of the granular flow, even if the energy redistribution accompanying the fragmentation is ignored. However, there is evidence that grain fragmentation during fault rupture is accompanied by intense (GPa-scale), high-frequency (kHz to MHz) stresses (Reches & Dewers, 2005); the resulting fragmented debris (called fault gouge) is very similar to that found in rock avalanche deposits (fractally distributed to tens-of-nanometer sizes, angular, containing agglomerates of finer grains; McSaveney & Davies, 2007), so it is reasonable to infer similar stresses in rock avalanches. These stresses arise as kinetic energy derived from the grain flow is slowly transformed to elastic strain energy in deforming grains in confined shear; upon grain fragmentation the stored elastic strain energy is released instantaneously as elastic body-waves in the fragments, resulting in additional intergranular stresses that affect motion. These stresses have been measured in grain fragmentation (e.g., Carpinteri et al., 2012; Schiavi et al., 2011).

Details of grain fragmentation are difficult to observe and quantify in rock avalanches and faults, for obvious reasons; nevertheless, the similar nature of the products, both in the field and in laboratory tests (Kolzenberg, Russell, & Kennedy, 2013), require that the suggested energy transformation and intense stresses are both real.

There is also evidence that when fragmentation occurs, localization of shear can be intense, as at the base of the Flims deposit in Switzerland (Figure 20), but even in this situation shearing still occurs in the overlying material.

Open in new tab

Figure 20. Basal contact of the Flims rock avalanche, Switzerland; sheared limestone lies above the contact (white intensely fragmented limestone), which overlies fluvial material. The deposit is distal.

Source: Photo by Anja Dufresne.

The Full Story

Integrating all of the above considerations, rock avalanche motion comprises the gravity-driven shearing flow of debris ranging in size from tens of meters (in the less-fragmenting carapace) to a few tens of nanometers. Fragmentation during the flow continuously produces fragments down to submicrometer-scale, many of the finest of which immediately agglomerate to form larger (but still silt-sized or finer) grains due to the propensity of freshly created rock surface to form molecular bonds with other rock in the immediate vicinity. Many fine fragments escape from the avalanche to form dust clouds reminiscent of pyroclastic flows, which preclude direct observation of the granular flow beneath the cloud.

Although fragmentation and its associated stresses appear to dominate the mechanics of the motion, dynamic and kinetic segregation still occur, and while fragmentation occurs throughout the moving mass, perhaps only a small percentage of grains is fragmenting at any one instant, so much of the motion is occurring as normal dense granular flow but affected by the high-intensity wave field generated by fragmenting grains.

Shearing occurs throughout the depth of the flow but is probably greatest at the base because strain rate is greatest there. Basal shearing may be intense enough to entrain substrate materials but not necessarily so even when moving over weak materials.

This overall picture is likely to vary locally due to varying basal topography and roughness, local water inputs due to the debris overrunning streams and lakes, and the fact that rock avalanches may result from retrogressive edifice failure so may comprise multiple moving debris masses close together in space and time.

Regional Sediment Supply

Although large rock avalanches occur relatively rarely in any given location, a large database of various types of landslides has been built up over the years that allows some statistical analysis. For example, Malamud, Turcotte, Guzzetti, and Reichenbach (2004) analyzed data from a wide variety of landslides in a range of environments and with a range of triggers and found that the probability density distributions of landslide area all collapsed onto a well-defined log-log plot. Korup and Clague (2009) further analyzed these and other data, finding that “fewer and larger landslides mobilize substantial fractions of the total debris volume, thus dominating the volumetric production rate of sediment.” While application of this conclusion to all landslides remains tentative due to incompleteness of databases, it has important geomorphological implications if valid, and there is no prima facie reason that it should not apply also to rock avalanches.

The major implication is that the larger landslides that occur in a mountainous region—which in non-volcanic regions will be mainly rock avalanches—supply more sediment to the landscape than do smaller mass movements. Hence, the sedimentology of the larger events will tend to dominate the whole of the sediment supply to the receiving landscape. This is likely to apply even in glacially dominated regions, because much of the sediment cycled through glaciers is sourced in mass movements that fall onto glaciers (Zemp, Kääb, Hoelzle, & Haeberli, 2005), rather than from subglacial erosion of bedrock (while mountains surrounding valley glaciers are generally being denuded at >1 mm a^-1, subglacial valley deepening occurs over a much smaller area and rarely reaches 1 mm a^-1; Harbor, 2013).

A previous section emphasized the large volumes of extremely fine (submicrometer) sediment that are generated by the rapid, intense shear occurring under confined conditions in rock avalanches. The only other common phenomena that create large volumes of extremely fine grains are earthquakes, but these fines are restricted to the fault rupture zones, so the supply of earthquake-generated fines to the landscape is restricted to the surface traces of active faults and hence is very small compared to rock avalanches. Thus rock avalanches appear to be the dominant source of submicrometer debris to landscapes. Again, glaciers are traditionally considered a major source of the finest landscape materials, but closer study indicates that the finest sediment grains produced by glacial erosion of bedrock are usually larger than 1 micrometer in size (e.g., Hooke & Iverson, 1995; Hubbard, Sharp, Willis, Nielsen, & Smart, 1995), and glacial sediments are commonly referred to as “silt-sized” (>2 μ‎m: e.g., Haldorsen, 1981). Considering that fragmentation in rock avalanches takes place at high stresses and high strain rates, while comminution beneath glaciers takes place at low stresses (because of the buoyancy of ice and the corresponding reduction in direct stress at a glacier base in a deep valley, in which a high water table is to be expected) and very low strain rates, it is reasonable to suggest that rock avalanches will generate a much higher proportion of ultra-fine sediments than glacial erosion.

Another characteristic of rock-avalanche–generated fines is the presence of sand- to clay-sized agglomerates comprising many smaller particles (Reznichenko et al., 2012). The agglomerates are long-lived in stationary deposits; Reznichenko et al. (2012) identified many of them in the 10,000-year old Waiho Loop terminal moraine in New Zealand, a temperate maritime environment that receives precipitation of ~8,000 mm a^-1.

Valley Long Profiles and Sediment Storage

A rock avalanche that falls into a mountain valley can, if large enough relative to the valley cross section, completely block the valley to form a landslide dam. A lake then forms behind the dam because the river flow is blocked; when the lake level reaches and overtops the dam crest, erosion of the downstream face of the dam by overflowing water often causes the dam to fail catastrophically sooner or later, generating a dambreak flood (e.g., Costa & Schuster, 1988). However some landslide dams do not fail in the short term; instead fluvial sediment builds up in the lake and eventually infills it completely, forming a flat area in the valley immediately upstream of the rock avalanche deposit. This gives many mountain valleys a characteristically stepped longitudinal profile (Korup, 2006). Hewitt, Clague, and Orwin (2008) describe a number of situations of this type, and Hewitt, Gosse, and Clague (2011) outline the impact of such events on landscape evolution in the Karakoram Himalaya. Korup, Densmore, and Schlunegger (2010) carry out the same analysis for other ranges; the role of rock avalanches in causing large volumes of sediment to be retained for long periods in mountain valleys is clear in all cases, as is the potential to significantly affect long-term landscape evolution.

Extreme Sediment Supply: Disturbance Regime Landscapes

Landslide dams resulting from rock avalanches often fail soon after their emplacement, long before they become infilled with sediment; in addition to this often devastating dambreak flood, which may remove only some tens of percent of the dam volume, major geomorphic change and damage can result from the longer-term reworking downstream of the remaining rock avalanche sediment. Following a severe mountain earthquake, river systems in a region may experience severe, widespread, and long-term floodplain aggradation (e.g., Robinson & Davies, 2013). There is evidence (e.g., Wells & Goff, 2007) that coastal dune systems in the southwest of the South Island, New Zealand, developed several decades after the earthquake that generated the sediment pulse causing the dunes to build, hence this is the time-scale on which severe river disturbance may be expected following a rock avalanche. Davies and Korup (2007) found that the Poerua River, Westland, New Zealand, resurfaced its upper fan over about a decade following a large rock avalanche a few kilometers upstream and inferred that on decadal time-scales the river would re-incise into the upper fan, which is thus not in equilibrium with the long-term drivers of landscape development. Hewitt et al. (2008) extend this concept to whole mountain landscapes, terming them “disturbance-regime” landscapes.

Rock Avalanches Onto Glaciers: Implications for Glacial Paleoclimatology

In mountains that still host valley glaciers, rock avalanches are likely to fall onto and run out over glacier surfaces (Deline, Hewitt, Reznichenko, & Shugar, 2015a). Rock avalanche deposits emplaced onto ice tend to spread much more, both laterally and longitudinally, than those emplaced onto earth materials, resulting in deposits only a few meters thick; hence a rock avalanche of 10⁷ m³ volume could cover an area of several square kilometers. If such a deposit should emplace onto the ablation zone of a glacier, the debris cover would drastically reduce the melting of ice beneath it, because the thermal mass of the debris would prevent diurnally varying solar energy from moving through the debris to the ice surface (Reznichenko, Davies, Shulmeister, & McSaveney, 2010); basal ablation (if any) would still occur but is a small proportion of total ablation from a debris-free glacier. If the debris covers a substantial proportion of the total ablation zone, the reduced ablation could cause the glacier to advance. During this advance the debris cover would be advected toward the terminus, because the ice-surface velocity reduces toward (and over) the terminus, so the debris-covered area will progressively reduce and the advance would eventually halt. The remaining debris-covered area would then stagnate and melt slowly, depositing an ablation moraine in front of the terminus (Figure 21, left), while the active terminus re-established itself at the pre–rock-avalanche position (assuming steady climate). The prominence of the moraine obviously depends on the volume of the rock avalanche relative to the size of the glacier.

Open in new tab

Figure 21. (Left) Small terminal moraines at Classen Glacier, New Zealand, showing sharp ridge with ablation moraine to its rear. Source: Photo by Stefan Winkler. (Right) The 5-km-long Waiho Loop moraine, Franz Josef glacier; this is the result of a supra-glacial rock avalanche about 10,000 years ago.

Source: Photo by George Denton.

Rock avalanche debris advecting to the advancing terminus would be dumped over it; if this dumping was sufficiently rapid it could block the advance, causing the terminal ice to thicken and allowing the terminal dump to increase in height as ice continues to move towards the stationary terminus. Thus a well-developed terminal moraine can result from a rock-avalanche–driven glacier advance. Figure 21 (left) shows such a moraine at the Classen Glacier, New Zealand, with the characteristic ablation moraine to its rear; also shown (Figure 21, right) is the renowned Waiho Loop terminal moraine of the Franz Josef glacier. This has long been the textbook archetype of a terminal moraine and was until 2008 associated with the Younger Dryas glaciation; in fact, this moraine was the main evidence that this glacial period extended from the northern hemisphere to the southern hemisphere. However, Tovar, Shulmeister, and Davies (2008) and Reznichenko et al. (2012) demonstrated that the Loop is composed predominantly of rock avalanche debris (angular, dominated by sandstone lithology that outcrops close to the highest part of the catchment, and containing plentiful agglomerates); and subsequently Alexander, Davies, and Shulmeister (2014) explained why and how the Loop formed where it did. The Loop has a volume of about 2 × 10⁸ m³, so the causative rock avalanche could have covered an area of about 40 km² (if the deposit was of the order of 5 m deep)—effectively the whole of the Franz Josef glacier ablation zone at the time. This reinterpretation necessitated a substantial revision of the glacial paleoclimatology of the mid-Holocene and of global climate teleconnections (Shulmeister, Davies, Evans, Hyatt, & Tovar, 2009). The ongoing and general implication is that existing moraine-based paleoclimate chronologies need to be examined for the possible presence of rock avalanche debris in the dated moraines, and future paleoclimate campaigns need to be designed with the possibility of rock-avalanche–generated moraines in mind (Kirkbride & Winkler, 2012).

A comparatively rare, but catastrophic, consequence of a rock avalanche falling onto a glacier can be a rock–snow–ice avalanche (Sosio, 2015). An excellent example is the 2002 Kolka event in the Caucasas (e.g., Huggel et al., 2005) in which a rock avalanche fell from Mount Dzhimarai-Khokh onto the Kolka Glacier; the impact of the rock avalanche caused the glacier to detach from its bed, and the large volume of detached ice became incorporated with the rock debris and formed a large composite mudflow. This flowed for about 19 km down the valley, killing about 100 people. A recent investigation into a large hummocky deposit in the Alai valley, Kyrgyzstan, revealed that, rather than being a late glacial moraine, it was a 28-km-long rock avalanche deposit; however, its extreme mobility was inferred to have been influenced by incorporation of glacier ice from below the rock avalanche source on Pik Lenin (Robinson, Davies, Reznichenko, & De Pascale, 2015).

Climate Change and Rock Avalanches

In recent years an usually large number of rock avalanches, triggered neither by earthquake nor rainfall, has fallen in glaciated—and therefore, in the current global warming (post–Little Ice Age) climate, deglaciating—mountains. Besides the six large events already referred to in New Zealand (Mt. Cook, 1991; two from Mt. Fletcher, 1991; Mt. Adams, 1999; Young River, 2008; Mt. Haast, 2013), there have also occurred the Saldim Peak rock avalanche, Nepal, 2017; the Nuugaatsiaq landslide, Greenland, 2017; the Lamplugh Glacier rock avalanche, Alaska, 2016; the Tyndall Glacier rock avalanche, Alakska, 2016; the Ferebee rock avalanche, Alaska, 2014; the La Perouse rock avalanche, Alaska, 2014; the Mt. Jarvis rock avalanche, Alaska, 2013; and the Mt. Lituya rock avalanche, Alaska, 2012. This five-year sequence appears extraordinary, even in the light of recent dramatic improvements in remote detection of such events via the earthquakes that their detachment and fall cause (e.g., Stark, Wolovick, & Ekstrom, 2012).

A number of investigations (e.g., Allen, Cox, & Owens, 2011; Deline, Broccolato, Moetzli, Ravanel, & Tamburini, 2013; Deline et al., 2015b; Haeberli, Schaub, & Huggel, 2017; Huggel, 2009; Krautblatter, Funk, & Günzel, 2013) have tested the hypothesis that the (apparent) increase in frequency of rock avalanches is the result of increased temperatures in the source areas by way of melting of permafrost. While the topic is complicated by the simultaneous occurrence of permafrost degradation and glacial downwasting and some of the data are equivocal, there appears to be a strong possibility that climatic warming corresponds to an increase in rock avalanche activity. For example, the Piz Cengalo rock avalanche referred to earlier has been associated with permafrost melting. By contrast, the anomalously high frequency of rock avalanches in Europe and elsewhere and in the mid-Holocene (e.g., Prager, Zangerl, Brandner, & Patzelt, 2007; Zerathe, Lebourg, Braucher, & Bourlés, 2014) has been associated with crustal stresses and increased seismicity following the last deglaciation (e.g., Ballantyne & Stone, 2013) rather than with any notable mid-Holocene change in climate.

Summary

Once it is appreciated that rock avalanches are the dominant source of sediment from denudation of mountains, their far-reaching geomorphological significance is unsurprising. Nevertheless, this broad significance has only recently started to become widely appreciated. For this reason alone, the processes and sedimentology of rock avalanches deserve greater attention from earth scientists. This attention has hitherto focussed mainly on the long-runout problem, but greater attention to the processes that generate the astonishing quantity of ultra-fine fragments—many of which then form agglomerates—is of equal or greater significance to the earth sciences in general.

Anticipation

Obviously nothing can be done to affect the behavior of a rock avalanche once it has detached from its parent edifice. Only if the location of this detachment can be identified in advance can measures be taken to reduce the impact of the event when it occurs. There are geomorphic signatures that seem to be associated with rock avalanches (e.g., Chigira et al., 2013); the stepped upper-slope profile (Figure 14) may indicate a future rock avalanche, preconditioned by seismic activity and triggered by either rainfall or earthquake.

In general, a large rock slope failure will, unless it is earthquake-triggered, release precursory rockfalls over the months to years preceding the eventual catastrophic collapse; thus the location can be studied and the volume of rock involved in, and perhaps even the approximate timing of, the collapse can be inferred—albeit with considerable potential errors (e.g., the Åknes landslide, Norway; Jaboyodeff et al., 2011).

Robinson, Davies, Wilson, and Orchiston (2016) developed a methodology for mapping the relative spatial susceptibility of specific 60 m² pixels to coseismic landsliding in a scenario (Alpine fault) earthquake in the South Island of New Zealand, but this would of course need to be reworked for every conceivable earthquake in order to provide a generalized spatial coseismic landslide susceptibility map. If such maps could be based on probabilistic seismic hazard distributions, then more general susceptibility distributions could be generated.

Detection

Even a rock large avalanche may be unrecorded in a vast, sparsely populated mountain range, and this in itself poses potential hazards to society because the consequential hazards—such as dambreak floods in locations far downstream—cannot then be anticipated. Recently, however, Goran Ekstrom and Colin Stark of Lamont-Doherty Earth Observatory have developed a system (Ekstrom & Stark, 2013) of monitoring seismological records to detect those generated by far distant rock avalanches. Although used previously (e.g., McSaveney, 2002; Moretti et al., 2012), the technique can now provide information on the location, volume, and dynamics of rock avalanches long before their deposits are seen by human eyes, and indirect impacts can perhaps be anticipated.

Direct Impacts

A rock avalanche impacting a town or village is devastating, as the events at Goldau and Elm referred to in the Introduction demonstrated. Such catastrophes are however (fortunately) extremely rare; nevertheless a large rock avalanche affecting any highly developed location in the future could cause enormous disruption, notwithstanding the fact that it would perhaps be anticipated due to detection of increasing precursory rockfalls. Even if the event were anticipated, the socio-political difficulties of organizing an evacuation—especially an urgent one—would be enormous, and the total costs of the event, including rebuilding, would be a severe fiscal challenge to any nation.

Barth (2014) illustrates a concerning situation of this type in New Zealand, where the township of Franz Josef Glacier lies immediately below a steep hillslope with a prominent bench near the top, very similar to that characterizing the slope alongside the Cascade rock avalanche source area (Figure 14), and again with the Alpine fault running along the base of the slope (Figure 22). If this should fail as a rock avalanche, the town could be obliterated, and during the 8-month-long tourist season this could impact thousands of people. This being a coseismic rock avalanche, no warning would be feasible except for the earthquake itself, which provides insufficient warning time for any remediation or evacuation to be effective. A back-of-the-envelope risk assessment of this situation is informative: Since deglaciation (prior to which the area was covered in ice to the level of the top of the steep slope; Barrell, 2011) there have been about 60 Alpine fault earthquakes, but the slope has not yet collapsed, so the probability of it collapsing in the next earthquake may be of the order of, say, 1%. The next earthquake on this fault has an annual probability of occurrence of about 2% (Cochran et al., 2017), so the annual probability of a coseismic rock avalanche is about 2 × 10⁻⁴. Even if only 100 lives are lost in this event, Figure 23 shows that the risk is well into the “unacceptable” range (Hungr, Clague, Morgenstern, VanDine, & Stadel, 2016). Fortunately there is a feasible remedy at Franz Josef—to relocate the town so that it is well out of the likely deposit area—but it remains to be seen whether this is societally and politically acceptable.

Open in new tab

Figure 22. Steep hillslope overlooking Franz Josef Glacier Township (lower center). Arrows indicate linear slope break similar to that in Figure 14 B = bulging mid-slope; chain dashed line = Alpine fault.

Source: Adapted from Barth (2014).

Open in new tab

Figure 23. Risk categories for landslides; approximate magnitude of Franz Josef rock avalanche risk indicated by red circle. ALARP, as low as realistically practicable.

Source: Hungr et al. (2016).

Indirect Impacts