Introduction

The frequency of large landslides in mountain cryospheric settings may be increasing globally (Huggel et al. 2010; Deline et al. 2015). These events present major hazards and contribute to changes in the dynamics of rock slope-glacier systems (e.g., Reznichenko et al. 2011; Shugar et al. 2012; Jamieson et al. 2015). Because of volumes typically over 106 m3, variable mechanical properties that change dynamically during runout, and interaction with runout path materials, large landslides (regardless of geological settings) have complex emplacement dynamics. The large number of emplacement hypotheses still debated in the scientific literature reflects this complexity (cf. Legros 2002; Davies and McSaveney 2012). Supraglacial landslides have slightly longer runouts (for a given volume) and are much thinner than their counterparts in other environments (Fig. 1; Evans and Clague 1988; Delaney and Evans 2014). The incorporation of snow (mixing, melting) and the interaction with glacial ice at the basal contact most likely contribute to the development of unique surface features and increased mobility.

Fig. 1
figure 1

Volume (V) versus runout (L) of long-runout rock avalanches: small black circles are rock avalanches over deformable substrates; large white circles (within the area outlined by an ellipse) are supraglacial rock avalanches; and triangles are debris flows

Full size image

Since direct observations of landslide emplacement are hindered by their hazardous nature and by large dust clouds that envelope their travel paths, surface features of the resulting deposits are used instead to reconstruct their motion sequences and to investigate their emplacement dynamics. These features include megablocks (meter- to decameter-sized blocks), faults (e.g., van Wyk de Vries et al. 2001; Coe et al. 2016), hummocks, longitudinal or transverse ridges, and flowbands (e.g., Clavero et al. 2002; Dufresne and Davies 2009; Paguican et al. 2014). Hummocks are circular to slightly elongate mounds with length-to-width ratios (L/W) of 1–2 and length-to-height ratios (L/H) typically below 10 (Dufresne and Davies 2009). Ridges and flowbands are both elongate surface features; ridges stand relatively high (up to 10 s m) above the surrounding debris (L/W > 2, L/H > 10), whereas flowbands are much narrower and longer (L/W ~ 20, L/H in the 100 s) and typically only a few meters higher than the surrounding debris (Dufresne and Davies 2009). Flowbands are typical for high-velocity mass movements (such as specific volcanic debris avalanches or pyroclastic flows) and in landslides where basal friction is low (e.g., travel over snow or glacial ice/snow). They appear to be intrinsic tendencies of all granular materials in motion and do not require dynamic internal processes, such as fragmentation, to form (e.g., Pouliquen et al. 1997; Aranson et al. 2006; cf. Dufresne and Davies 2009). In supraglacial landslides, flowbands are particularly long and prominent, and in the absence of complex substrate and topographic conditions, can provide key insights into the dynamics of granular mass movements.

Indirect observations of landslide emplacement have become available in recent years through the analysis of the seismic signals generated by large landslides upon interaction with the ground (e.g., Kanamori and Given 1982; Dammeier et al. 2011; Moretti et al. 2012; Allstadt 2013). Forces and momenta at the source, when and where major changes take place in landslide momentum along the runout path, and the mass of the material involved can be estimated from seismic data (e.g., Ekström and Stark 2013; Hancox and Thomson 2013; Hibert et al. 2014, 2015, 2017).

Whereas sizes, distributions, and the formation of hummocks have been studied in great detail (e.g., Paguican et al. 2014), and flowbands have often been described and analyzed geometrically (e.g., Dufresne and Davies 2009), only a few systematic studies combining field data of structure and sedimentology, high-resolution topographic data, and continuous seismic data exist on recent, unmodified deposits (Jiskoot 2011; Guthrie et al. 2012; Coe et al. 2016). The combination of these techniques is a valuable proxy to direct observation.

In this study, we analyze the surface features and the emplacement dynamics of the 28 June 2016 rock avalanche on Lamplugh Glacier in Glacier Bay National Park and Preserve, SE Alaska. This remote rock avalanche was detected as a seismic event by the long-period event detector at the Lamont Doherty Earth Observatory (cf. Ekström and Stark 2013). The processes and dynamics of this event are reconstructed in high spatio-temporal detail through temporally continuous seismic data combined with a high-resolution digital elevation model (DEM) and field observations gathered during site visits 6 days after the slope failure occurred.

Geological setting

Geological setting

The Lamplugh rock avalanche is located in Glacier Bay National Park and Preserve (GBNP) in southeast Alaska (Fig. 2). The park is named after Glacier Bay proper, which is a fjord that resulted from post Little Ice Age (LIA) deglaciation (e.g., Connor et al. 2009). The glaciated peninsula, where the Lamplugh rock avalanche occurred, is located along the tectonically active plate boundary between the Pacific and North American plates. The peninsula is undergoing rapid uplift, primarily due to rebound from post LIA deglaciation (Larsen et al. 2005).

Fig. 2
figure 2

a Regional setting of the Lamplugh rock avalanche within Glacier Bay National Park and Preserve in southeast Alaska, showing the area covered by aerial photogrammetry survey on 4 July 2016 (blue polygon) and the rock avalanche deposit. b Satellite image of the deposit collected on 16 July 2016 by DigitalGlobe Inc.; yellow polygons outline areas investigated during ground traverses

Full size image

Bedrock exposed in GBNP is complex and consists of Paleozoic and Mesozoic accretionary terranes containing Tertiary sedimentary, intrusive, and volcanic rocks (Brew et al. 1978; Wilson et al. 2015). Mapping by Gruber (2012a; 2012b) indicates that permafrost is likely in bedrock at high elevations within the study area. Mountain peaks and ridgelines in these high elevation areas are the source areas for frequent rock avalanches (e.g., Geertsema 2012). At least 25 rock avalanches occurred since 1984; though not all were large events (Bessette-Kirton and Coe 2016; including one in the Mount Fairweather area post 2016). The size and mobility of these rock avalanches appear to increase with time (Coe et al. 2018). The Lamplugh rock avalanche is as of yet the largest in the historical record for GBNP (Bessette-Kirton and Coe 2016; Bessette-Kirton et al. 2018). The park could be viewed as one of the bellwether locations for rock avalanches caused by warming due to climate change.

The 2016 Lamplugh rock avalanche

The Lamplugh rock avalanche measures 10.5 km in length, with a maximum width of ~ 3.5 km and a deposit area of 21 km2. The total source and deposit volumes were estimated to be 52 ± 2 Mm3 and 70 ± 8 Mm3, respectively (Bessette-Kirton 2017; Bessette-Kirton et al. 2018). It originated from a north-facing bedrock ridge at an elevation of ~ 2180 m above sea level (asl). The slope was partially covered by snow and ice, and had an initial slope of ~ 51°. From the pre-failure summit, the rock mass descended ~ 900 m to a bench at the foot of the mountain, and ran out onto the Lamplugh Glacier, which has an average surface slope of ~ 1.3°. The toe of the main deposit is located at an elevation of ~ 560 m asl and is approximately 8.6 km inland of Lamplugh Glacier’s terminus in Glacier Bay. The ratio of fall height to travel distance (H/L) of the Lamplugh rock avalanche (1620/10,500 m) is 0.15. A second rock slope failure occurred 2 h after the main event and formed the herein called secondary deposit on top of the proximal rock avalanche deposit of the main event.

Detailed examination of the source area was impossible due to the extreme height and steepness, and because of continued rockfall activity during the time of field work. However, regional mapping by Wilson et al. (2015) shows that the mountain consists of sedimentary and metamorphic rocks of the Triassic to Cretaceous Kelp Bay Group, including phyllite, quartzite, greenschist, greenstone, greywacke, and greywacke semi-schist. The trigger for this event cannot be pinned to a single factor. Seismicity, however, can be ruled out since no strong earthquakes occurred in the hours before the slope failed. Increased melting at higher altitudes and consequent permafrost degradation was likely a factor in conditioning this slope for failure (Coe et al. 2018), since the winter 2015/16 and the spring 2016 were the warmest in recorded history (Walsh et al. 2017).

Methods and data

Field and laboratory work

Field work was carried out on 4–5 July 2016, 6 days after the failure, and an aerial photogrammetry survey was conducted on the clear day of 4 July. Glacier conditions allowed landing only in the upper, i.e., proximal reaches of the deposit since snow melt began to expose the crevassed ice surface and made landing too risky in the days following our investigations. From the landing point at the right-lateral margin, we traversed the rock avalanche deposit to the most proximal runup location and hiked towards the source area (boxes in Fig. 2b). Motion detectors (PlotWatcher) were installed outside the rockfall area of the source as well as on a small ridge in the proximal depositional area for 24 h, recording one frame every second to document short-term deposit changes. Several pits were dug across the deposit to expose the interior and base, samples of the rock avalanche were taken, and dust on the surrounding ice/snow surface was collected in the proximal reaches. The rock samples were weighed before and after drying at room temperature to estimate their moisture content. Grain size distributions of the dust samples were analyzed by laser diffractometry at the department of Physical Geography, RWTH-Aachen University, Germany.

Photogrammetric data acquisition and processing

We performed an airborne photogrammetry survey of the Lamplugh rock avalanche area on 4 July 2016. Digital photographic data were acquired using a fixed-wing airborne platform with flight lines and photograph frequency that provided 60% side lap and 80% end lap coverage, with an average flight height of 1040 m asl. The airborne photogrammetry survey covered 149 km2, resulting in 1992 photos that were used to generate 0.19-m resolution orthoimagery and a 0.38-m resolution DEM.

A Nikon D800 camera with AF-Nikkor 24 mm f/1.4D lens was used to collect 36.2-megapixel JPEG photographs (7360 × 4912 pixels per image) which were compressed for optimal quality. During the aerial survey, the photograph coordinates were determined using a Cirrus Digital Systems intervalometer that linked the camera shutter release to a Trimble R7 Global Navigation Satellite System (GNSS) receiver, with a Zephyr 2 antenna, that had a 5-Hz background sampling frequency. The camera was mounted inside the aircraft with the GPS antenna positioned 0.96 m above the camera. The GPS antenna offset was corrected during GPS post-processing to solve for the camera coordinates. A Trimble R8 model 4 GNSS receiver was deployed approximately 75 km southeast of the survey area and was used to differentially correct the aerial survey GNSS data.

Aerial survey GNSS data were processed using GrafNav GNSS Post-Processing Software and post-processing kinematic (PPK) methods. The base station position was corrected using the National Geodetic Survey OPUS with the IGS08 (EPOCH:2015.6162) solution. Both aerial and base GNSS data were collected and processed in WGS84 (G1674) using the WGS84 ellipsoid. Standard deviations for the airborne GNSS camera positions were less than 0.06 m in both horizontal and vertical directions.

Aerial stereo-photographs were imported into the commercially available structure from motion (SfM) software (Agisoft 2018) to align aerial photos, edit the sparse point cloud, optimize the bundle block adjustment, construct the dense point cloud and triangulated irregular network geometry, and generate and export the DEM and the natural color RGB orthoimagery.

Digital mapping and analysis

Structural features were mapped in a geographic information software system at scales between 1:8000 and 1:3000 on the high-resolution 0.38-m DEM that was acquired on 4 July 2016. Orthophotos (0.19-m resolution) from the same day were overlaid and set at 50% transparency to aid in structure identification.

We used a 15 June 2016 2.0-m resolution DEM generated from high-resolution (~ 0.5 m) WorldView satellite stereo imagery to represent the pre-event surface (Bessette-Kirton and Coe 2018). The pre- and post-event DEMs were co-registered and differenced in order to map areas of accumulation and depletion, and to estimate the thickness of the rock avalanche deposit. Since a direct calculation of the thickness of the rock avalanche deposit in depleted (scoured) areas is not possible, following Bessette-Kirton et al. (2018), we assumed the average thickness of surrounding deposits to represent the average thickness of the deposits in the depleted areas.

To better highlight and enhance surface structures, morphologic features, and their relative formation timing with respect to one another, we created a grid of surface roughness from the 0.38-m DEM; surface roughness derived from slope aspect proved most useful. To create this aspect surface roughness grid, we first removed an overall, first-order surface trend from the 0.38-m DEM and then calculated an aspect value for each cell. Lastly, we calculated an aspect roughness value for each cell in the aspect grid as the focal sum of a 3 × 3 kernel surrounding each cell.

Seismic data and analysis

The seismic waves generated by the Lamplugh rock avalanche were recorded by several short-period and broadband stations from the Alaska Regional Network (AK), National Tsunami Warning System (AT), Canadian National Seismograph Network (CN), and the USArray Transportable Array (TA) networks, located at distances ranging from 122 km to more than 400 km from the source. The first seismic wave arrival was observed at the closest station (AK.BESE) at 16:21:03 UTC.

The Lamplugh Glacier rock avalanche generated many large-amplitude long-period surface waves that can be used to perform an inversion of the Landslide Force History (LFH; e.g., Ekström and Stark 2013). The inversion of the force imparted on the Earth by the rock avalanche was conducted using the parametric approach proposed by Ekström and Stark (2013). The source was parameterized in terms of 10 sub-sources, each with a duration of 20 s and separated in time by 10 s. We assume that all the forces act on the Earth at the same location—a reasonable approximation given the wavelengths of the long-period seismic waves that are modelled. The total duration is thus estimated to be 110 s. Each sub-source represents an impulse (force multiplied by duration) acting on the Earth in an arbitrary direction. The 10 impulses have to add up to 0, as no net force is acting on the Earth. The inversion finds the amplitudes and directions of the 10 impulses, as well as a best time and location for the centroid of the forces.

We equate the negative of the force acting on the Earth with the force (mass multiplied by acceleration) of the landslide mass (e.g., Ekström and Stark 2013; Hibert et al. 2014, 2015). This quantity is then integrated once over time to get the momentum of the slide (mass multiplied by velocity), and a second time to get the mass multiplied by the displacement of the center of mass.

In addition to long-period seismic wave inversion, short-period seismic wave processing can provide insightful information on the dynamics of the event (e.g., Hibert et al. 2014, 2015). Hibert et al. (2017) have shown that for large landslides, the amplitude of the envelope of the seismic signals in the 3–10 Hz frequency band is controlled by the momentum of the center of mass. Hence, a complementary estimate of the momentum and other dynamic parameters can be inferred from the short-period seismic signal corrected for propagation and attenuation effects.

The Lamplugh rock avalanche deposit(s)

Large (> 106 m3) long-runout rock avalanches (RA) spread highly fragmented debris over several square kilometers at velocities of tens of meters per second. They result from the sudden failure of rock slopes, and the initially relatively intact rock mass breaks up during failure and runout. Their mobility is higher than what can be expected from simple frictional considerations, hence the prefix “long-runout” (cf. Hsü 1975).

Depositional areas

The Lamplugh rock avalanche deposit is divided into five depositional areas based on morphological features and associated processes (Fig. 3):

  1. 1.

    the main deposit featuring prominent flowbands,

  2. 2.

    runup along the left-lateral margin,

  3. 3.

    fallback,

  4. 4.

    raised margins, and

  5. 5.

    a secondary deposit from a later failure on top of the proximal main deposit.

Fig. 3
figure 3

The five depositional areas of the Lamplugh rock avalanche: main deposit (gray), raised margin (orange), runup (light blue), fallback (dark blue), and the secondary deposit (pink). The locations referenced by other figures in this paper are shown in white boxes; those not indicate here are located within the yellow boxes in Fig. 2

Full size image

Prominent long flowbands define the main deposit along its entire 10.5-km length. Four areas of substantial runup (29–151 m) and fallback along the left-lateral margin provide data for velocity calculations. The margins, apart from the fallback areas, are raised and have unique surface roughness signatures that are different from roughness signatures within fallback areas. A secondary deposit with a length of 3.8 km from a later failure sits on top of the proximal main deposit.

Deposit composition and thickness

The Lamplugh rock avalanche deposit is composed of fragmented source rock and of snow and ice derived from both the source and deposit areas. Photographic evidence suggests that only a very small fraction (~ 1%) of the source material is snow and ice, which is consistent with adjacent slopes where snow accumulation is hampered by extremely steep terrain. An additional 17–24% of the deposit material is entrained snow and ice (Bessette-Kirton et al. 2018). Incorporation of snow and ice from the runout path through bulldozing is most evident along the raised margins (Fig. 4a). The ice-snow mixture forms diamicts (Fig. 4b; similar to the nearby La Perouse rock avalanche (Geertsema, unpubl. data) and elsewhere, e.g., Yamasaki et al. (2014)), boulders with internal shear planes (Fig. 4c), occurs as large broken chunks (Fig. 4d), or as entire rafted blocks of several 100 m3 inherited from the source.

Fig. 4
figure 4

a Snow and ice at the lateral margin; geologist for scale. b Ice diamict. c Shear plane in ice “boulder.” d Rock and ice surface in the deposit interior; pilot for scale in ellipse

Full size image

The thickness of the landslide debris varies between a few centimeters over ice and where surface clast sizes are smaller than boulders (Fig. 5a, b), to several meters at locations that are covered by boulders and megablocks (Fig. 5c). Trenches dug in several locations in the deposit interior revealed that chaotic snow with a clear basal contact to structured snow forms the lower parts of debris even though these appeared to be composed of rock debris when viewed from the surface alone. Differencing pre- and post-slide DEMs shows the variation in deposit thickness wherein areas of fallback, distal bulldozing, and the proximal rockfall accumulation are the thickest parts (Fig. 6a), and flowbands occur in areas of more thinly spread deposition (Fig. 6b).

Fig. 5
figure 5

Local variations in debris thickness range from a few centimeters of fine-grained debris over almost undisturbed snow layers (a; shovel is about 0.5 m long, and b; small tool is 20 cm long) to several meters of ice and snow with grain sizes up to several meters across (c); geologist for scale (photo credit for (c): Colin Stark)

Full size image
Fig. 6
figure 6

(a) Variations in deposit thickness. (b) Areas of accumulation and depletion relative to pre-slide topographic surface

Full size image

Field observations suggest that color differences in the deposit areas that are predominantly composed of source rock, with a lighter-colored deposit center and darker margins (Fig. 7a), are mainly due to differences in moisture content rather than in composition. Even smaller-scale variations in color are due to moisture content; i.e., samples of different color collected in the field are of the same lithological composition and turn the same color after drying. Moisture loss was ~ 5.5 wt.-%, and larger “clasts” disintegrated during drying and handling (Fig. 7b, c). Dust on the glacier surface outside the rock avalanche deposit is sparse, and the two samples recovered during the field visit are dominated by grain sizes between 0.63 and 63.0 μm, with only 3–12% coarser than 63.0 μm, and less than 2% smaller than 0.2 μm.

Fig. 7
figure 7

a Delineation of the relatively dry central part (white polygon) from the outer, more saturated debris (darker area outside the white polygon) of the rock avalanche deposit. b The grain sizes appear overall coarser in the field than after handling (c) and drying. (c) Larger clasts turn out to be agglomerates (yellow circles in b and c) of finer material. The entire sample turned to a uniform light gray color after drying (not shown)

Full size image

Longitudinal surface features

Flowbands describe morphological features within a deposit (cf. introduction), whereas lobes specify geometrical features of tongue-like extensions beyond the main deposit.

At the Lamplugh deposit, lobes formed all along the unobstructed deposit margins. Some extend like fingers a few hundred meters farther than their direct neighbors (Fig. 8a, b). Flowbands within each lobe are (sub-) parallel to the lobe’s direction of motion. The lobes themselves, however, can diverge by up to ~ 75° from the main event’s trajectory (which is due NNE at 008°). Tracing the flowbands within each lobe back into the main deposit gives a rough estimate of where each lobe started. In some cases, this “point of origin” can be traced back by over a kilometer (Fig. 8b, e).

Fig. 8
figure 8

Prominent surface structures of the Lamplugh rock avalanche. a Flowbands (represented by the furrows between them); note increase in flowband density distribution in the distal half-third of the deposit. b Detail of flowbands curving with the respective local debris trajectory in lateral lobes; note how flowbands are parallel within each individual lobe, and how offsets along flowbands only occur between lobes, but not within lobes. The yellow lines indicate starting points of lobes and flowbands. The one with the broad starting “point” has a very different morphology than “regular” lobes. The deformation features, particularly at its toe, indicate continued movement after the adjacent debris had come to rest. c Outline of the secondary deposit on top the proximal main deposit. d Detail of the secondary deposit with strike-slip faults between its own individual lobes. e Surface roughness map of one of the “runaway” lobes (cf. b), which continued sliding after the main deposit had come to rest. This lobe has very different roughness signatures, and areas of collision into already stationary debris and small changes in motion direction in response to the collision are visible

Full size image

Two lobes stand out from the rest by distinctly different surface textures (Fig. 8b, e). Both extend farther beyond the main rock avalanche mass than other lobes and moved independently even after the main mass came to rest (evidenced by collisional deformation of the adjacent stationary rock avalanche debris). They began to diverge from their own initial trajectory and spread laterally as soon as their toes passed the surrounding rock avalanche debris. Their starting points extend ~ 1.8 km (the right-lateral lobe; Fig. 8b) and ~ 1.1 km (the distal lobe; Fig. 8a) into the main deposit. Furthermore, their starting “points” are much broader than those of the other “regular” lobes (Fig. 8b). A more chaotic surface texture and more prominent transverse ridges, which are particularly evident on surface roughness maps (Fig. 8e), characterize these two lobes.

Flowbands are the most prominent structural surface features on the Lamplugh rock avalanche deposit. They extend through all depositional areas, but are most abundant in the distal ~ 3 km and along the lateral margins (Fig. 8a; black lines are the furrows between the flowbands, which presented clearer mapping targets). They are straight in the deposit interior (i.e., along the main movement trajectory) and curve along the lateral margins where they apparently reacted to changes in underlying topography and followed the steepest local gradient, and/or curved due to lateral debris spreading, and/or were perturbed by neighboring lobes. We mapped 1156 furrows defining the flowbands with lengths between 21 and 1700 m (median of 231 m; Supplement-A). Within individual lobes, about 25% of flowbands display offsets between their neighboring flowbands of up to a maximum of ~ 10 m (the other 75% show no apparent offsets; this is out of a preliminary sample of 100 flowbands). Between neighboring lobes, however, strike-slip faults with offsets of up to 400 m developed between the bounding flowbands to allow differential movement needed to accommodate lobe formation.

The strike-slip faults were additionally highlighted (in the days after emplacement, i.e., before snow melt set in) through offsets in the lighter-and-darker-colored debris and/or bands of bulldozed snow (Fig. 9). Within the secondary deposit (Fig. 8c, d), the offsets between lobes are smaller (on the order of tens of meters) than in the main deposit with the parts towards the deposit center having moved farther relative to the secondary deposit’s lateral margins, similar to the main deposit.

Fig. 9
figure 9

a Offset along a runup and fallback zone where the main rock avalanche body moved past and dragged some of the material from the fallback area (arrow represents ~ 80 m ground distance; this location is marked R4 in Fig. 16c). b Strike-slip fault between two distal lobes; the relative displacement is clearly visible between the transverse ridges of entrained snow. The arrows represent ~ 50–55 m ground distance (the location is the same as the one in Fig. 11)

Full size image

Transverse surface features

Transverse and oblique features are linear to curvilinear structures that are oriented at ~ 45–90° to the main motion trajectory (and thereby to the flowbands). These features are best identified along the deposit margin where bands of compression and extension alternate (Figs. 9b, 10a, and 11). Some extensional features entail a stretch of exposed ice in their wake and bulldozed transverse ridges at their prow (Fig. 10a). Adding all areas of exposed ice along a profile in the deposit center reveals that deposit elongation through late-stage “passive/brittle” (Fig. 10b) extension (as opposed to “dynamic” deposit thinning) amounts to 2.2% of the total deposit length (38 extensional features with a sum of 228 m total extensional length; assuming similar extension of the main and the secondary event).

Fig. 10
figure 10

Evidence of late-stage brittle extension of rock avalanche debris in the a distal and b proximal deposit areas. Arrows point in motion direction of the debris

Full size image
Fig. 11
figure 11

a Orthophoto at 50% transparency overlaid on hillshade. The arrows point to a pre-slide glacial ridge that is “printed through” the deposit, but did not affect the formation or direction of flowbands. b Oblique aerial view of the same location

Full size image

Alternating bands of rock and ice/snow characterize the raised margins (Figs. 9b and 11) and imply entrainment through bulldozing and thrust faulting around the rock avalanche rims. This phenomenon is well known for many rock and volcanic debris avalanches (e.g., Johnson 1978; Evans and Clague 1998; Belousov et al. 1999; van Wyk de Vries et al. 2001; Clavero et al. 2004; Abdrakhmatov and Strom 2006; Strom and Abdrakhmatov 2018) as well as from laboratory analogue models (e.g., Shea and van Wyk de Vries 2008; Dufresne 2012). Strike-slip faults along lobe boundaries offset transverse bands of entrained snow/ice, indicating that differential motion between individual lobes occurred late during emplacement, i.e., during the last phase of deceleration and after entrainment/bulldozing (this concept is explored in more detail below). Interestingly, only a few of the compressional features disrupt flowbands within individual lobes (Figs. 9b and 11).

Pre-slide runout path structures

Ridges present on the pre-slide glacier surface influenced neither rock avalanche motion nor the development of surface features. In fact, some flowbands traverse glacial ridges (arrows in Fig. 11) at an oblique angle and remain unaffected in both orientation and width.

We also examined the distribution and geometry of crevasses on the surface of the Lamplugh Glacier in pre-event imagery (2 October 2015; Supplement-B) to determine their potential influence on deposit structures. Visible differences in the sizes and directions of extensional features within the rock avalanche deposit indicate that these features were indeed formed through primary emplacement processes and were not influenced by crevasses in the underlying surface.

Megablocks

Large boulders (i.e., megablocks: meter- to decameter-sized blocks) are scattered, apparently randomly, throughout the entire deposit, and the deposit hence lacks the coherent network of large boulders that defines a carapace (cf. Davies and McSaveney 2012). They often lie within small depressions which possibly formed by ice/snow melting underneath the block. Many of these megablocks are capped with finer rock debris and ice (see Fig. 5c). We mapped the distribution of those larger than ~ 5–10 m diameter/long axis. Most of these 565 megablocks occur to the east of the central trajectory (Supplement-C).

Features inherited from source

Several cliffs of up to a few hundred meters in length that expose ice and snow within the deposit stand several meters high above the adjacent rock avalanche surface (Fig. 12a, b). They are identical in appearance to cliffs in the thick ice-snow cover that drapes the slopes along the flanks of the failure scarps (Fig. 12c). We hence interpret them to have been transported ice blocks, rafting most likely, that came to rest seemingly intact. One rafting ice block was transported along the full 10.5 km runout length and came to rest at the deposit toe between two lobes (Fig. 12a).

Fig. 12
figure 12

a Large snow-ice package that was rafted 10 km to the distal end of the rock avalanche and deposited seemingly intact; the yellow arrow points at an approximately 17-m-high cliff (see Fig. 2 for location). b Cliff within the proximal rock avalanche deposit consists almost entirely of snow and ice with a thin veneer of rock debris. c Snow-ice package at the right-lateral margin of the source area (photo credit: Colin Stark)

Full size image

Rapid deposit changes

The Lamplugh rock avalanche debris changes rapidly in this dynamic environment. Within the 6 days between emplacement (Fig. 13a) and site investigations (Fig. 13b), much of the surficial ice and snow had disappeared, meltwater ponds formed, and melting of the lower ice layers constantly caused small-scale movement and reorganization of the overlying debris (e.g., rocks tumbling from melting ice/snow-dominated ridges).

Fig. 13
figure 13

a Photograph from the day of emplacement (with permission from Paul Swanson, Mountain Flying Service LLC, Alaska). b Six days later, much of the snow had already melted; note ponds formed on the debris surface (white arrows). Although the main features are still recognizable, many details which the snow had highlighted are already less distinct

Full size image

Therefore, as ice and snow melt and saturated debris dries out and ablates, the identification of some structural features becomes more challenging since the color differences between the materials initially highlighted features such as faults, debris stretching, or material entrainment much more clearly.

Landsat images from 29 July 2016, 2 September 2017, and 3 July 2018 show that during the first and second year following emplacement, the rock avalanche deposit moved in the down-glacier direction at average rates of 97 and 70 m/year, respectively (Fig. 14). A lack of velocity data for the Lamplugh Glacier precludes us from determining whether these changes primarily resulted from continued deformation of the rock avalanche deposit (e.g., Uhlmann et al. 2013), were related to glacier movement, or whether the rock avalanche changed glacier dynamics as reported for other sites (e.g., Shulmeister et al. 2009; Vacco et al. 2010; Reznichenko et al. 2011; Shugar et al. 2012; Stark et al. 2012; Jamieson et al. 2015).

Fig. 14
figure 14

Changes of the distal deposit margin between 29 July 2016 and 3 July 2018 overlain on a pan-sharpened (15-m resolution) Landsat image collected on 3 July 2018. The deposit margin locations in 2016 and 2017 were also delineated from Landsat imagery

Full size image

Within the failure scar, rockfall activity remained constant during the days of site investigation. The PlotWatcher camera outside the rockfall area recorded that rockfall activity was lower during the night than during daytime for the 24-h recording period.

Rock avalanche dynamics

From seismic and morphological data, we identified two emplacement stages (simply referred to as Stage-1 and Stage-2) during the main rock avalanche event which covered the entire 10.5 km runout distance. These two stages and their dynamic implications are explored below. Two hours after the main event, a secondary slope failure occurred, resulting in a secondary deposit on top of the proximal main deposit; no separate emplacement stages were identified for this secondary event.

Seismic signal analysis

Detection and signal features

The first arrival of the seismic waves generated by the Lamplugh rock avalanche was detected at the closest station (AK.BESE, 122 km from the source) at approximately 16:21 UTC on 28 June 2016 (Fig. 15a). From this station, we observed a signal with a duration of 320 s (Fig. 15c). The signal is characterized by an emergent onset and a spindle-shape envelope (from 16:21:00 to 16:24:00 at AK.BESE, Fig. 15c), as classically observed for landslide-generated seismic signals. This first part of the signal corresponds to the first emplacement stage: Stage-1. At most stations, a peculiar long coda can be observed (from 16:24:00 to 16:26:20 at AK.BESE, Fig. 15c). This second part of the signal corresponds to Stage-2. Whereas the first part (Stage-1) of the signal exhibits strong energy in the low frequency band (0.001–1 Hz), the coda (Stage-2) is characterized by the absence of long-period waves and by higher energy in the highest frequency band (3–10 Hz, Fig. 15c).

Fig. 15
figure 15

a Seismic signals of the Lamplugh Glacier rock avalanche (main event) recorded at seven seismic stations from the AK and the TA networks. b Seismic signals generated by a secondary rock avalanche 2 h after the main event (a). c Seismic signal recorded at the station AK.BESE, filtered in three frequency bands. The first and second stages of the seismic signal (and emplacement) of the main event are indicated by the yellow and the blue bars, respectively

Full size image

Following the main event, several other seismic signals with characteristics suggesting landslides and larger rockfalls have been observed. The strongest of those signals occurred at approximately 18:21 UTC on 28 June 2016 (Fig. 15b). The chronology of the detection of first arrivals by the stations of the local seismic network suggests that the seismic waves originated from the same source area as that of the main event.

Long-period processing

The inferred Landslide Force History (LFH; Fig. 16a) shows an initial upwards force, corresponding to unloading of the Earth as the slide mass starts accelerating downslope. The largest initial horizontal force is to the south, indicating that the slide mass accelerated to the north. The location of the centroid is estimated to be 58.57° N, 136.97° W. The maximum of the modulus of the inverted forces is 2.8 × 1011 N and the maximum of the modulus of momentum is 7.8 × 1012 N/s. Knowing the mass allows inferring the acceleration, the velocity, and the trajectory of the center of mass. With a mass of 141 × 109 kg calculated from the source volume and an average rock density of 2720 kg/m3 (Mankhemthong et al. 2013), the maximum acceleration yielded by the LFH inversion is 2.0 m/s2, the maximum velocity is 54.9 m/s, and the runout distance (of the center of mass) is 2853 m; the full inverted trajectory is presented on Fig. 16b.

Fig. 16
figure 16

a Components of the force imparted by the rock avalanche on the Earth inverted from the long-period seismic waves. b Envelope of the seismic signal recorded at BESE (gray), velocity profile derived from the LFH inversion (blue) and from the short-period envelope analysis (red), and velocities inferred from geomorphological observations (black diamond—R1 to R5; with R5 representing the hypothetical velocity at the distal rock avalanche point assuming linear deceleration). c Hillshade of the source and deposit along with the points at which the geomorphological velocities have been inferred (R1 to R4), the trajectory deduced from the LFH inversion (black line along deposit center)

Full size image

Short-period processing

To estimate the amplitude of the short-period waves at the source, we computed a simple attenuation model taking into account inelastic attenuation and loss of energy related to geometrical spreading (Hibert et al. 2017) from the signals recorded at 16 stations, with distance to the source ranging from 122 to 425 km. This yields a value of the maximum amplitude at the source of 1.73 ± 0.58 × 10−3 m/s. Using the scaling law between the maximum amplitude at the source and the maximum momentum proposed by Hibert et al. (2017), we estimate the maximum momentum between 1.2 and 2.0 × 1013 N/s.

Assuming that the amplitude of the short-period seismic signal envelope observed at the closest stations is not dominated by propagation effects and is essentially controlled by the intrinsic source characteristics, which is supported by the observation that the envelope shape does not vary much between stations, we can use the envelope of the seismic signal recorded by the closest station (AK.BESE) to determine a velocity profile of the center of mass. To do so, the amplitude of the signal envelope recorded at BESE is normalized by the maximum amplitude at the source and converted to momentum following the scaling law proposed in Hibert et al. (2017). Knowing the mass, the velocity at each time can be derived from the resulting momentum curve. By integrating this velocity profile, we can infer the total runout distance. We find values ranging from 6.9 to 12.8 km. The velocity profile that corresponds to the observed runout distance (8985 m starting from the base of the source slope) is obtained with a maximum velocity of 84.0 m/s (Fig. 16b).

Runup and velocity

The rock avalanche debris ran part way up the slopes on its left-lateral margin at four locations with runup heights decreasing with increasing distance travelled (R1–R4 in Table 1 and Fig. 16c). From these data, the debris velocity is estimated using a simplified approach wherein only conversion of kinetic to potential energy is considered in v = (2gh)0.5, where v is the velocity, g the gravitational constant, and h the height of runup against topography (proposed first by Chow 1959, and later established as a common approach for landslide velocity estimates; e.g., Jibson et al. 2006; Iverson et al. 2016). All velocity values were plotted against distance and extrapolated to the distal end of the deposit (Table 1 and Supplement-D). This extrapolation was implemented to estimate the travel time of the distal debris of the rock avalanche beyond the last runup (R4), and thereby calculate the total runout duration. The starting point for the runout calculations is the base of the source scarp (“point zero” in Fig. 16c), i.e., the moment that the rock mass started to spread as a rock avalanche after the initial mass collapse and fracturing.

Table 1 Runup, velocity, and runout time
Full size table

The peak velocity of 54.4 m/s estimated from runup for the Lamplugh event lies within the range of those published for other long-runout landslides on and off glaciers (e.g., Hewitt 1988; McSaveney 2002; Sosio et al. 2008; Jiskoot 2011). The Lamplugh velocities, however, may only represent minimum values since they are based on measurements along the deposit margins. A velocity cross-profile with a relative increase in velocity towards the debris center may be deduced from offsets along strike-slip faults and farther travel of lobes in the center relative to the margin. However, these faults might also be due to differential lateral deceleration as debris along the margins stopped first while the center still moved forward at a similar velocity (eyewitness accounts and videos from other mass movements around the world support this latter interpretation).

The peak velocity derived from LFH inversion is similar to the peak velocity estimated from runup (R1). However, the LFH-derived velocities are lower during the first rapid deceleration (from 16:21:10 to 16:22:10; Fig. 16b) of Stage-1. The velocities derived from the short-period seismic signal analysis fit well with the geomorphological velocities during the first deceleration (of Stage-1), but are slightly overestimated during the acceleration (of Stage-1), and underestimated during the steady flow (Stage-2; Fig. 16b). Several assumptions could explain those discrepancies. First, it is important to note that while the seismic signal analyses provide an estimate of the velocity of the center of mass, the geomorphological velocities are deduced from runup heights on the edges of the flow. Local effects from the topography underlying parts of the flow and the full complexity of the propagation dynamics cannot be retrieved through seismic signal analysis. Concomitantly, the velocities derived from geomorphological observations might not capture entirely the global dynamics of the flow as they provide local measurements only. This might alter the estimation of the times when R1 to R5 were reached. Finally, the scaling law between the maximum momentum and the maximum amplitude of short-period seismic waves at the source is still ill-constrained (Hibert et al. 2017) and brings strong uncertainties on the velocity profile deduced from the envelope of the short-period seismic signal, which might partly explain the difference with the LFH- and SP-derived kinematic quantities.

Reconstructed emplacement dynamics

The 28 June 2016 rock avalanche deposited a sheet of rock, ice, and snow over 10.5 km distance from source down Lamplugh Glacier. Morphological and seismic data indicate that the deposit was produced by two separate rock slope failures, 2 h apart. The first failure at 16:21 produced the main deposit (21.0 km2), moving over the low-friction glacial snow and ice surface along its full length. The second rock slope failure at 18:22 spread debris over an area of 3.4 km2 onto the high-friction rock avalanche debris of the main deposit, producing a secondary deposit in the proximal deposition area. Both events must have lacked full body turbulence since megablocks on the debris surface are capped by smaller-grained debris, which would have been worked into the flow had they rotated during emplacement.

Our data suggest that, after failure and descend down the rockslope, the main event moved in two distinct emplacement stages (Figs. 15c and 17). Seismic analyses of the long-period and the short-period waves yield insights into these stages. The LFH inferred from the long-period waves inversion suggests a first stage (Stage-1) of propagation where the center of mass has undergone a fast northward acceleration and then a deceleration during a cycle lasting approximately 2 min and over a distance of close to 3 km (of the center of mass). After Stage-1, only short-period waves are observed (Fig. 15c). This suggests, together with the velocity profile inferred from short-period analysis (Fig. 16b), that Stage-2 of the propagation was dominated by low-velocity sliding. This very slow deceleration stage, and hence the absence of strong acceleration, would cause no long-period seismic waves to be generated. The lack of long-period seismic waves precludes the use of the LFH inversion to reconstruct the dynamics of the flow during this second stage, which might explain why the runout distance and the duration given by the LFH is shorter than the observed runout and the one inferred from the short-period seismic wave analysis.

Fig. 17
figure 17

a Runout distance and morphological features of Stage-1 (granular flow) of the main rock avalanche event at 16:21. b Stage-2 (sliding) of the main event. c Later continued sliding of two individual lobes. d The secondary rock avalanche at 18:22

Full size image

Clear long-period seismic signals during Stage-1 indicate strong interaction of the rock avalanche debris with the ground. This stage lasted for approximately 100 s. In this time, the front of the spreading debris reached ~ 5 km (accounting for uncertainty of the initial collapse duration) from the base of the source slope (Figs. 16c and 17a). We interpret the strong interaction with the ground as a granular flow powered by internal processes, such as grain collisions and comminution, acoustic fluidization (e.g., Melosh 1979; Collins and Melosh 2003), “granular temperature” (Campbell 1989), and/or dynamic fragmentation (e.g., Davies 1982; Davies and McSaveney 2009). The exact physical mechanisms behind the generation of seismic signals during rock avalanche emplacement are still under investigation (Hibert et al. 2017; Bachelet et al. 2018; Farin et al. 2018). The seismic signal at a given time in a granular flow is the sum of the impulses imparted by the grains on the Earth. The impulse of each grain could be controlled by all the processes listed above, which in turn will be controlled by the overall dynamics of the flow. Thus, at the first order, the global dynamics of the flow will control the amplitude of the short-period seismic signal. The emplacement dynamics of rock avalanches are still lively debated in the literature today, over 130 years after the first hypotheses were proposed by Buss and Heim (1881).

During Stage-2, low friction was so low and deceleration so slow that rock avalanche motion was essentially aseismic at long periods. It coincides with a change in surface texture to a much higher number and density distribution of flowbands (Fig. 17b). Hence, rock avalanche motion during Stage-2 was essentially by low-velocity sliding on the low basal friction surface. In addition, the dry area in the deposit center becomes much narrower beyond the roughly 5 km transition point between the two stages (Fig. 7a). We interpret this as increased entrainment of glacial snow and ice, which saturated the debris. The aseismic mass propagation suggests that any dynamic internal avalanche processes apparently ceased; at least to below the detection limit. A video of the front of the Mt. Dixon supraglacial rock avalanche in New Zealand (Hancox and Thomson 2013; https://www.youtube.com/watch?v=E28_3uj9K0g) documents this process of “passive sliding” of rock avalanche debris on snow, and entrainment and bulldozing of snow from the runout path at the flow front (cf. Dunning et al. 2015). At Lamplugh, this is additionally observed along the right-lateral margin where flowband density increases within the individual lobes, and surface roughness, debris thickness, and snow entrainment also increase.

That flowbands are intrinsic tendencies of all granular materials in motion and do not require dynamic internal processes, such as fragmentation, to form is suggested by numerous laboratory experiments of non-fragmenting granular flows (e.g., Pouliquen et al. 1997; Aranson et al. 2006) and empirical research (cf. Dufresne and Davies 2009). The combination of seismic data, morphological observations, eyewitness documentations, and granular motion theory as well as experiments point to our conclusion of Stage-1 “active emplacement” (fragmenting granular flow) and Stage-2 “passive sliding” (steady state).

Conclusions

The combination of field mapping, high-resolution DEM and orthophoto acquisition, and seismic signal analyses carried out in this study proved invaluable in understanding the dynamics of the Lamplugh supraglacial landslide. This landslide shows two emplacement stages: (i) granular flow with strong interaction of the landslide debris with the ground (“active or dynamic emplacement”), and (ii) “passive sliding” in the distal roughly third of the runout length. Stage-1 is characterized by clear long-period seismic signals, its deposit is thin and stretched with a dominant dry and flat area in the center, and has narrow raised margins. Stage-2 is aseismic at long-periods, and features a deposit with a higher density of flowbands and prominent raised margins of entrained/bulldozed snow. The formation of flowbands is still not understood. We observe strike-slip faults with up to 400 m offset between lobes, but less significant (< 10 m) to no offsets between flowbands within individual lobes. The higher apparent mobility of supraglacial landslides (relative to their counterparts in other runout environments) may be explained by the two-stage model developed herein. The proposed “passive sliding” of Stage-2 is in agreement with eye-witness accounts and (e.g. at the Mt. Dixon slide in New Zealand). In general, field surveys of landslides in rapidly changing glacial environments should be carried out right after emplacement to capture all information recorded in the deposit before melting and/or cover by snowfall and/or glacier motion obscure and distort features.