Snow cornices develop along mountain ridges, edges of plateaus, and marked inflections in topography throughout regions with seasonal and permanent snow cover. Despite the recognized hazard posed by cornices in mountainous locations, limited modern research on cornice dynamics exists and accurately forecasting cornice failure continues to be problematic. Cornice failures and associated cornice fall avalanches comprise a majority of observed avalanche activity and endanger human life and infrastructure annually near Longyearbyen in central Svalbard, Norway. In this work, we monitored the seasonal development of the cornices along the plateaus near Longyearbyen with a terrestrial laser scanner (TLS) during the 2016–2017 and 2017–2018 winter seasons. The spatial resolution at which
we acquired snow surface data with TLS enabled us to observe and quantify
changes to the cornice systems in detail not previously achieved. We focused primarily on the evolution and failure of the lower cornice surfaces where accessibility has precluded previous research. We measured cornice accretion rates in excess of 10 mm h
Snow cornices are overhanging projections of snow that form due to the deposition of wind-transported snow in the lee of ridgelines or sharp slope inflections (Montagne et al., 1968; Seligman, 1936). Cornices have attracted interest for their hydrologic implications (e.g., Anderton et al., 2004) and as agents of geomorphic change in periglacial environments (Eckerstorfer et al., 2013; Humlum et al., 2007), but they are perhaps best recognized as a snow and avalanche hazard in mountainous terrain (Montagne et al., 1968; Vogel et al., 2012). Cornices pose an avalanche hazard when they fail either as a full cornice failure with the entire cornice detaching from the ground or as a partial failure with a smaller cornice mass separating from the rest of the cornice. The detached cornice blocks travel downslope under the influence of gravity and become a cornice fall avalanche by entraining loose surface snow or triggering a snow slab on the slope below (e.g., Vogel et al., 2012). In ski areas or where cornices and cornice fall avalanches endanger infrastructure, both explosives (Farizy, 2013; McCarty et al., 1986) and structural defenses (e.g., Montagne et al., 1968) are employed operationally to mitigate cornice hazards. Most cornice-related fatalities, however, occur in recreational backcountry settings and result from the victim's weight triggering cornice failure.
Despite the well-recognized hazards and operational challenges associated with cornices and cornice fall avalanches, specific cornice research is relatively scarce. Early cornice studies summarized by Vogel et al. (2012) focused on qualitative descriptions of cornice formation processes and resulting cornice structures (e.g., Montagne et al., 1968; Seligman, 1936). Later studies investigated mechanisms by which individual snow crystals adhere during cornice accretion (Latham and Montagne, 1970), the physical snow characteristics at various structural locations on individual cornices (Naruse et al., 1985), and the specific interactions between wind-drifted snow and cornice morphology during cornice formation (Kobayashi et al., 1988).
Recent work has refined the conceptual model of seasonal cornice dynamics
established by these earlier studies primarily by employing time-lapse
photography to examine cornice responses to the meteorological factors
controlling the development and failure of cornices (Munroe, 2018; van Herwijnen and Fierz, 2014; Vogel et al., 2012). Vogel et al. (2012)
observed cornice processes over two winter seasons on a single mountain slope in central Svalbard and proposed a conceptual model of seasonal cornice dynamics including cornice accretion, deformation, and failure. Their results indicated cornice accretion occurs during or immediately following winter storms with wind speeds in excess of 10 m s
Later work in an alpine setting also linked cornice accretion to strong winds during or soon after a snowfall and found the SNOWPACK wind drift index correlated well with cornice width estimates (van Herwijnen and Fierz, 2014). No cornice failures or cornice fall avalanches were observed in this study, however. Munroe (2018) used time-lapse photography to observe the growth and repeated failure of a cornice in Utah, USA. He also found cornice accretion to primarily coincide with periods of snowdrift. He divided the 19 cornice fall avalanches observed in his study into two distinct groups: snow-caused cornice fall avalanches where failure primarily resulted from snow loading on the cornice and temperature-caused avalanches where failure was related to rapid temperature increases presumably leading to destabilization of the cornice through the loss of snow strength.
We build upon the observational understanding and conceptual model of seasonal cornice dynamics established in these previous works by monitoring cornice systems in Longyeardalen – including one site previously examined by Vogel et al. (2012) – with a terrestrial laser scanner (TLS). TLS – or ground-based lidar (light detection and ranging) – is an active remote sensing technology with documented applications for observing and monitoring various slope processes and hazards including landslides (Jaboyedoff et al., 2012; Prokop and Panholzer, 2009), coastal cliff erosion (e.g., Caputo et al., 2018), and rock slope instability (Abellán et al., 2014). TLS is being increasingly employed in snow and avalanche research to map snow depth and snow depth change (e.g., Deems et al., 2013; Fey et al., 2019; Prokop, 2008; Schirmer et al., 2011). Other specific snow-related applications include quantifying snow drift processes to verify physical models (Mott et al., 2011; Schön et al., 2015; Vionnet et al., 2014), observing avalanche activity to calibrate dynamic avalanche models (Prokop et al., 2015), assisting avalanche control operations (Deems et al., 2013), and planning and designing snow fences that limit hazardous snow accumulation in avalanche release areas (Prokop and Procter, 2016).
We monitored cornice accretion, deformation, failure, and associated cornice
fall avalanche activity near Longyearbyen, Svalbard, with TLS technology over
two winter seasons (2016–2017 and 2017–2018). To our knowledge TLS has not
been employed to specifically monitor cornice dynamics, so our primary
objectives are to use the high-spatial-resolution snow surface data acquired
via TLS to
demonstrate the utility of TLS to observe cornice processes; observe and quantify cornice accretion, deformation, failure, and associated cornice fall avalanches and link these processes to their controlling meteorological factors; use our findings to provide suggestions for forecasting cornice fall
avalanches in this and other locations threatened by cornices.
The present study focuses on the cornices forming above Longyeardalen (“the Longyear valley”) in central Svalbard (Fig. 1). Longyeardalen is a glacially sculpted, U-shaped valley with a northeast–southwest-oriented axis running approximately 3 km from the termini of two small mountain glaciers to a fjord. The Gruvefjellet and Platåberget plateaus border Longyeardalen to the west and east, respectively, with Svalbard's administrative center, Longyearbyen, situated in the valley bottom. The Gruvefjellet and Platåberget slopes lie within the horizontally bedded, lower-Tertiary Van Mijenfjord Group of sandstones and shales (Major et al., 2001). Resistant strata within this group form the area's extensive plateau topography. The entire region is underlain by continuous permafrost ranging in thickness from 100 m near the coasts to over 500 m in the higher mountains (Humlum et al., 2003).
Overview of Longyeardalen and key locations, including automated
weather stations (AWSs), mentioned in the text. Contour lines in
We investigated seasonal cornice dynamics and cornice fall avalanches along
and under the Gruvefjellet and Platåberget plateau margins, respectively
(Fig. 2). The steep valley walls descending from the broad plateau summits
(approximately 450 m elevation) are characterized in their upper portions by
protruding resistant bedrock buttresses and transport couloirs incised by
fluvial and gravitational slope processes. The Gruvefjellet slope described in detail by Eckerstorfer et al. (2013) consists of a 50–70 m near-vertical bedrock cliff situated under the plateau margin and over a 40–50
Overview of the cornice systems and locations of the primary spatial data employed in this work. Panel
Central Svalbard's climate is cold and arid, with a mean annual air temperature of
The climate of Svalbard prohibits the growth of woody vegetation, and snow distribution across the landscape is thus strongly controlled by the wind (e.g., Jaedicke and Sandvik, 2002). Southeasterly winds generally prevail across the region's plateau mountains but often switch to westerly or southwesterly during winter storms and are frequently redirected along the major valley axes at lower elevations (Christiansen et al., 2013). Winter weather in central Svalbard fluctuates between extended periods of cold, stable high pressure punctuated by warm, wet low-pressure systems conveyed northwards along the North Atlantic cyclone track (Hanssen-Bauer et al., 1990; Rogers et al., 2005). This is reflected in the region's snow and avalanche climate, where the snowpack typically consists of persistent weak layers formed during high pressure interspersed with wind slabs or ice layers formed during snowstorms or rain-on-snow events (Eckerstorfer and Christiansen, 2011a). Avalanche activity here displays a strong topographical and meteorological control, with direct action slab avalanches clustered around winter storms and the region's plateaus serving as source areas for the extensive cornice systems that contribute to frequent cornice fall avalanches (Eckerstorfer and Christiansen, 2011b).
We obtained wind and air temperature data from the Gruvefjellet automated weather station (AWS), precipitation data from the Svalbard Airport AWS, and a limited time series of snow depth data from a pair of ultrasonic snow depth sensors placed in avalanche release areas on Gruvefjellet and Platåberget during the 2017–2018 winter season (Figs. 1 and 2). We defined the winter season as 1 December to 30 June for the purposes of this study. The Gruvefjellet AWS is located less than 500 m east of the Gruvefjellet cornice system at an elevation of 464 m and records hourly meteorological data. The Svalbard Airport AWS is situated approximately 5 km northwest of the study area at 28 m and is the only weather station in the region with long-term precipitation measurements.
As part of the installation of a network of automated snow monitoring stations in Longyeardalen (Prokop et al., 2018), we mounted two ultrasonic snow depth sensors in avalanche release areas under the cornice systems in autumn 2017. These sensors were located at 350 and 450 m elevation on Gruvefjellet and Platåberget, respectively (Fig. 2). We employed the Campbell Scientific SR50A ultrasonic distance sensor to measure snow depth at each location. The snow sensors began recording reliable snow depth data on 15 November 2017 and continued until the end of the 2017–2018 season at 10 min intervals.
We used a Riegl® Laser Measurement Systems VZ-6000 ultra-long-range terrestrial laser scanner to repeatedly scan the Gruvefjellet and Platåberget cornice systems throughout the 2016–2017 and 2017–2018 winter seasons. The VZ-6000's 1064 nm operating wavelength is particularly well-suited for measuring snow surfaces, while the high scanning speed and measurement range up to 6 km with a 30 kHz pulse repetition rate ensured adequate data acquisition capabilities across the study area in a variety of atmospheric conditions (Riegl® Laser Measurement Systems, 2019; Prokop, 2008).
We use data from 25 scans of Gruvefjellet and 22 scans of Platåberget during the duration of the study (Appendix A). Of these, one scan from Gruvefjellet and Platåberget each is a snow-free surface taken on 16 September 2016. For Gruvefjellet, we acquired usable snow surface data from 18 scans during the 2016–2017 season and seven scans during 2017–2018. We acquired 14 snow surface scans of Platåberget during 2016–2017 and seven scans during 2017–2018. The TLS was unfortunately damaged in late April 2018, and we were unable to acquire any scans after our final scan on 13 April 2018.
We preprocessed the raw point clouds in RiSCAN Pro, Riegl's proprietary
data processing software. We established a suite of ground control points on
both Platåberget and Gruvefjellet using a differential global positioning system (DGPS) which we used to georeference individual point clouds. We then aligned repeated snow-covered scans to the snow-free scans established in September 2016 using these ground control points and the “Multi-Station Adjustment” plugin in RiSCAN Pro following the approach outlined by Prokop and Panholzer (2009). We then manually filtered non-ground points or points above the snow surface. Finally, we applied an octree filter with a 0.10 m increment and exported to an
Visualization of the point cloud processing methods in CloudCompare. Panel
We imported individual point clouds into CloudCompare (CloudCompare, 2019) for further analyses (Fig. 3). To create 2-D cornice profile cross sections, we extracted point cloud profile sections along manually defined axes using the polyline extraction tool native to CloudCompare (Fig. 3c and d). This tool requires user-defined inputs for profile type, section thickness, and maximum edge length which we set to “both”, 0.6 and 0.2 m. We then manually edited and digitized the resulting shapefiles in the ArcScene 3-D Editing environment (ArcGIS 10.4.1) to create the vertical cornice profile schematics as 3-D shapefiles.
We calculated representative volumes for selected areas from both the
Platåberget and Gruvefjellet cornice systems using the “compute 2.5-D
volume” tool in CloudCompare. This tool computes the volume between two
2.5-D point clouds by rasterizing the point clouds to a specified grid size
and then computing volumes based on the differences in a specified projection direction between the rasterized values (Fig. 3e–h). In our case, we rasterized our point clouds to a 1 m grid and calculated horizontal distance differences along the “
We used the Multiscale Model-to-Model Cloud Comparison (M3C2) algorithm developed by Lague et al. (2013) and implemented as a plugin in CloudCompare to quantify changes in the cornices and snow surfaces on the slopes below in 3-D. The M3C2 algorithm allows for direct comparison of point clouds in 3-D and is specifically developed to handle 3-D differences and detect changes to complex surfaces where both vertical and horizontal changes exist (Lague et al., 2013). This functionality requires the user to input the following parameters: the normal scale, the projection scale, and the maximum depth (e.g., Lague et al., 2013; Watson et al., 2017). We selected a normal scale of 2 m oriented positively to the scan position (i.e., the normals “face” the scan position), a projection scale of 1 m, and a maximum depth of 10 m for all M3C2 calculations.
TLS-based snow surface measurement accuracy generally decreases with increasing distance from the scanner to the measured snow surface and is
affected by the manner in which the laser beam interacts with the snow surface, the local terrain characteristics, the stability of the scanner while scanning, and the quality of the scan data registration process (Fey et al., 2019; Hartzell et al., 2017; Prokop et al., 2008). The relative accuracy – the deviation between measurements of an unchanged surface taken under different measurement conditions – can be assessed to quantify uncertainties related to both registration errors and positional errors from the interaction of the laser beam with the surface (Fey et al., 2019; Prokop and Panholzer, 2009). We assessed relative accuracy for our data by measuring M3C2 distances between each snow-covered scan and the snow-free scan on a 10 m
We relied on snow and avalanche observations from Platåberget and Gruvefjellet from the Norwegian Water Resources and Energy Directorate's (NVE) online observation platform regObs (
We compare seasonal meteorological conditions (Fig. 4) with cross-sectional cornice profiles derived from eight scanned snow surfaces on Gruvefjellet and seven surfaces on Platåberget. We selected these profiles from a pool of 18 usable scans from Gruvefjellet and 14 from Platåberget (Appendix A) to represent key points in the development of the cornice systems.
Summary of the representative cornice volume progression and meteorological conditions for the 2016–2017 winter season. Wind speed and air temperature are daily averaged values from the Gruvefjellet AWS, and precipitation data are daily values from the Svalbard Airport AWS measured at 06:00 UTC. Shaded blue vertical bars indicate well-constrained cornice accretion periods for which we were able to calculate horizontal cornice accretion rates (Table 2). Shaded grey vertical bars indicate 48 h periods with observed noteworthy cornice fall avalanche activity (Table 1).
Small cornices had accumulated on Gruvefjellet by 2 December 2016. Maximum
horizontal cornice growth prior to this scan occurred in the vicinity of
profile GF2, where both vertical and horizontal cornice growth exceeded 1 m
from the edge of the plateau (Fig. 5). The representative cornice volume in
the vicinity of profile GF1 already approached 200 m
The 2-D cornice profiles showing cornice progression for selected scan dates throughout the 2016–2017 winter season. Each profile is labeled as it is referred to in the text and corresponds to the location and POV depicted in Fig. 2.
Heavy snowfall followed by strong westerly winds preceded several cornice fall avalanches on 21 January on Platåberget (Fig. 4, Table 1). Representative cornice volume on Platåberget nearly doubled from roughly
300 to over 600 m
Summary of avalanche cycles.
A major accretion event in mid-February 2017 followed several weeks of
unseasonably high temperatures at cornice elevation during early February
(Fig. 4). Locally heavy snowfall and strong easterly winds accompanying a
vigorous winter storm impacted the region between 19 and 21 February. Profile GF1's horizontal extension increased by nearly 3 m between the 17 and 24 February scans, resulting in horizontal accretion rates exceeding 15 mm h
Summary of well-constrained accretion events.
Representative volumes for both cornice systems gradually increased in the
following month, and profiles from 21 March 2017 show considerable rounding
and downslope creep of the cornices' leading edges in profiles GF1 and GF3
(Fig. 5). Cornices continued to grow on Platåberget, with horizontal
growth exceeding 2 m on portions of the PB1 and PB2 profiles and PB3's vertical extent increasing by over 2 m. The Platåberget cornices did not deform downslope to the same degree as the Gruvefjellet cornices during this time period. A representative volume decrease of over 500 m
We gathered seven scanned snow surfaces from both Gruvefjellet and Platåberget for the 2017–2018 season with which to compare to meteorological conditions. Cornice development during the 2017–2018 winter
season differed considerably from the 2016–2017 winter season despite relatively similar
seasonal meteorological conditions (Table 3). Gruvefjellet profiles from 15 December 2017 show over 5 m of horizontal cornice growth in all profiles,
and representative volume approached 1000 m
Summary of the representative cornice volumes and meteorological conditions for the 2017–2018 winter season. Wind speed and air temperature are daily averaged values from the Gruvefjellet AWS, precipitation data are daily values from the Svalbard Airport AWS measured at 06:00 UTC, and snow depths are daily averages from the snow sensors on Gruvefjellet and Platåfjellet. Shaded grey vertical bars indicate 48 h periods with observed noteworthy cornice fall avalanche activity (Table 2).
The 2-D cornice profiles showing cornice progression for 2017–2018 winter season scan dates. Each profile is labeled as it is referred to in the text and corresponds to the location and POV depicted in Fig. 2.
Seasonal summaries. All parameters are measured at the Gruvefjellet AWS except for precipitation, which is measured at the Svalbard Airport AWS.
Representative volume doubled on Platåberget between the 31 January and
22 February scans from 400 to 900 m
Snow depths increased by 0.20 and 0.28 m on Gruvefjellet and Platåfjellet, respectively, on 26 and 27 February 2018 as over 7 mm
precipitation was recorded at the airport (Fig. 7). A marked increase in
representative volume of 230 m
We documented three periods of cornice fall avalanche activity with TLS data
in April 2017. In the first, a small portion of the cornice between profiles GF1 and GF2 failed on 9 April following a period of precipitation falling as snow and easterly winds in excess of 10 m s
Meteorological summary of the April 2017 case study. Wind speed, wind direction, and air temperature are hourly values from the Gruvefjellet AWS, and precipitation data are daily values from the Svalbard Airport AWS measured at 06:00 UTC. Colored vertical lines in the time series indicate the scan timing corresponding to the profiles in Fig. 9. Vertical grey bars marked (a)–(c) correspond to 48 h time periods with noteworthy avalanche activity discussed in the text.
M3C2 distances displaying changes to the snow cover on Gruvefjellet between the 21 March and 25 April 2017 scans
Cornice profiles illustrating cornice dynamics during the April 2017 case study, with each profile labeled as it is referred to in the text. Dashed lines indicate interpolated data where overhanging cornice structure shadowed the snow surface from the TLS.
A warm winter storm accompanied by 4.5 mm of precipitation, southwesterly winds, and air temperatures approaching 0
A storm in mid-March 2018 punctuated a month of otherwise stable weather and
resulted in cornice fall avalanches on Gruvefjellet (Fig. 11a). From 15 to 19 March, 5.6 mm of precipitation accumulated at the airport AWS, snow depths at the Gruvefjellet sensor increased by a maximum of 18 cm while those at the Platåberget sensor decreased by approximately 0.25 m, and
strong winds blew from the ENE for 24 h on 17–18 March. Two large cornice failures on Gruvefjellet visible as strongly negative M3C2 distances near profile GF1 and slightly to the north (Fig. 12a, annotation 1) triggered avalanches on the slope below (Fig. 12a, annotation 2). Similar to the morphology observed in the April 2017 cornice fall avalanches, the failed cornice blocks stripped snow off the vertical rock face and created impact craters while entraining snow as they moved downslope. The cornice chunks from these cornice failures also remained intact throughout the event and ran further than the rest of the avalanche debris (Fig. 12a, annotation 3). A cornice block approximately 5 m in horizontal extension detached from the cornice represented by profile GF1, while a smaller (
Meteorological summary of the March 2018 case study. Wind speed, wind direction, and temperature are hourly values from the Gruvefjellet AWS, and precipitation data are daily values from the Svalbard Airport AWS measured at 06:00 UTC. Colored vertical lines in the time series indicate the scan timing corresponding to the profiles in Fig. 12, and the grey vertical bar annotated with (a) corresponds to the 48 h time period with noteworthy avalanche activity mentioned in the text.
M3C2 distances displaying changes to the snow cover on
Gruvefjellet
Cornice profiles illustrating cornice dynamics during the March 2018 case study, with each profile labeled as it is referred to in the text. Dashed lines indicate interpolated data where overhanging cornice structure shadowed the snow surface from the TLS.
TLS-derived cornice data from the 2016–2017 and 2017–2018 winter seasons provide quantitative reinforcement to the conceptual models of cornice dynamics developed in previous studies (e.g., Montagne et al., 1968; Vogel et al., 2012). In these models, cornices accrete through relatively discrete events and begin to deform under their own weight before either failing or melting away towards the end of the snow season.
Our data show cornices can rapidly accrete at any point in the snow season
given abundant snow available for wind transport, wind speeds sufficient to
mobilize surface snow, and wind directions oriented relatively perpendicular
to the ridgeline. We documented accretion rates in excess of 15 mm h
Following initial accretion, the cornices' leading edges begin to deform downslope. Deformation becomes more pronounced later in the season, presumably as increased air temperatures and solar radiation begin to warm the snow, decreasing the stiffness of the cornices and increasing creep (e.g., Schweizer et al., 2003). Further accretion events can then be superimposed on this deformation as the season progresses, with short accretion events interspersed by longer periods of downslope creep. This can be seen in the minor increases in horizontal extension and continued downslope deformation in GF1 and GF3 through the latter portion of the 2017–2018 season (Fig. 7). Cornice accretion and downslope deformation can also occur almost simultaneously with air temperatures approaching or even exceeding freezing at cornice level, as evidenced by the rapid accretion and downslope creep shown in profile PB1 for the 25 April–1 May scan interval (Fig. 10).
While meteorological conditions control the specific timing of cornice
accretion and downslope deformation, the underlying topography appears to
act as a fundamental control on cornice structure and seasonal cornice dynamics. The presence of the steep bedrock face directly beneath the
Gruvefjellet plateau margin limits the support provided by the underlying
topography compared to the gentler sloping Platåberget margin. The result is a more overhung cornice structure on Gruvefjellet, while Platåberget's topography allows for more slope-normal snow accumulation. Profiles PB1 and PB2 failed to develop cornices at all during the 2017–2018 winter season (Fig. 7). The presence of cornices with horizontal extension approaching 5 m in these locations during the 2016–2017 winter season (Fig. 5), however, shows the topography can support cornice development given the right meteorological conditions. Differences in meteorological conditions between the 2016–2017 and 2017–2018 winter seasons may provide a partial explanation for differing seasonal snow cover responses on Platåberget (Table 3). Winds in excess of 5 m s
Topography also seems to control the relative size of cornice failures. Vogel et al. (2012) describe a “geomorphologically determined sedimentary step approximately 3 m below the plateau that most likely acts as the cornice pivot point” on Gruvefjellet. This pivot point is most evident in profile GF1, where in both winter seasons the downslope creep of the overhung cornice beyond this pivot point ultimately became overburdened during an accretion event and caused the cornice to fail completely. The cornice represented by GF1 has the least topographic support and developed the most overhanging cornice structure of the specific cornices we investigated, and also failed completely both seasons. By contrast, the topographic support provided by Platåberget does not promote overhanging cornices to the same degree, instead promoting a thicker slope-normal snowpack which in itself supports the cornice structure. Here, observed cornice failures such as that shown in PB2 during the 25 April–1 May 2017 scan interval (Fig. 10) are limited to the recently accreted snow and did not involve the entire cornice mass. Similarly, profile GF2 failed in March 2018 within hours of profile GF1's full failure but involved a much smaller portion of the cornice predating the 2 March scan – potentially related to increased topographic support to this cornice relative to GF1 (Fig. 13).
Previous work has differentiated cornice fall avalanche types by the inferred mechanism of cornice failure – via either increased snow load from accretion or decreased snow strength in the cornice related to increased snow and air temperatures. Five of the six cornice fall avalanche events observed in this study coincided with winter storms leading to accretion just prior to cornice failure (Table 1). This is in contrast to previous findings from this location, where no cornice failures were observed in direct response to snow loading caused by a snowstorm (Vogel et al., 2012). The lone cornice fall avalanche event we cannot link to cornice accretion occurred in January 2018. This event coincided with heavy precipitation, but positive temperatures at the Gruvefjellet AWS and decreasing snow depths at the Gruvefjellet snow sensor indicate this precipitation fell as rain (Fig. 6). Our truncated TLS observation record in late spring 2018 unfortunately omits the May–June period found by Vogel et al. (2012) to be critical for air-temperature-induced cornice failures in this location, but observational records throughout this time do not indicate further cornice fall avalanches. Accretion's role in determining cornice failure is also reflected in the asynchronous timing of cornice failures on Gruvefjellet and Platåberget during our study. None of the observed avalanche events included activity on both Gruvefjellet and Platåberget simultaneously as would be expected with air-temperature-induced failures, with avalanches instead occurring only on the leeward aspect.
Observed cornice fall avalanche size appears to be controlled largely by the snow conditions in the underlying release area. Cornice fall avalanches on Gruvefjellet follow a pattern exemplified by the April 2017 case study in which the cornice fails and removes snow from the steep bedrock face below as it descends before impacting the release areas at the base of the cliff (e.g., Fig. 9a). The cornice block can then, depending on the snow conditions in the release area, entrain snow from its impact crater and the avalanche path below or trigger a larger slab avalanche. Cornice failures near profile GF3 in both April 2017 and March 2018 triggered small slab avalanches, but the majority of the avalanche debris resulted from entrainment as the cornice blocks bounced downslope.
Platåberget's topography promotes slightly different avalanche dynamics. The gentler slope at the plateau edge allows snow to accumulate directly under the cornices such that failed cornice masses land directly on the snow to be released as an avalanche. Release areas on Platåberget collect snow during accretion events much more efficiently than those on Gruvefjellet, where blowing snow mass losses due to suspension are promoted by the separation created between the cornices and the release areas by the bedrock cliff. Accumulation in the upper release areas on Platåberget coinciding with accretion events primes these locations for slab avalanche release with even small cornice failures. Relatively small cornice failures triggering larger slab avalanches on Platåberget in April 2017 resulted in magnitude avalanches (D2, R2–R3) comparable to those releasing from much larger cornice failures but less entrainable snow on Gruvefjellet in March 2018 (Figs. 9b and 13a).
Cornice fall avalanches are the most common avalanche type observed in the portion of central Svalbard surrounding our study area where the broad plateau summits and steep valley walls of Longyeardalen's topography are recurrent across the region (Eckerstorfer and Christiansen, 2011b). Cornice fall avalanches observed in this study thus represent processes occurring elsewhere throughout central Svalbard – and to a lesser extent other locations throughout the world – and provide an opportunity to reinforce existing forecasting frameworks with detailed cornice data. The conceptual model of avalanche hazard in North America treats cornice failure both as an individual avalanche problem to be considered by forecasters and as a potential trigger when assessing the likelihood of other avalanche types releasing in a given forecasting area and time period (Statham et al., 2018). Cornice fall avalanche hazard assessments should thus consider both the likelihood of cornice failure and the nature of the snow conditions in the release area to best judge cornice fall avalanche hazard. Our limited dataset, especially in the absence of multiple air-temperature-induced failures, is insufficient to make broad generalizations linking cornice failure type and resulting cornice fall avalanche activity. As a specific example, however, fairly widespread wind slab avalanche activity throughout the region accompanied each of the accretion-induced avalanche events observed in this work. The conditions leading to cornice accretion – strong winds and available snow for wind transport – also promote the development of wind slab problems. Thus, when conditions are favorable for cornice accretion and accretion-induced cornice failures, conditions are also favorable for the development of more widespread – and potentially more sensitive – slab avalanche problems. In this scenario, the chance of a cornice failure triggering a secondary slab avalanche would rise, subsequently amplifying the cornice fall avalanche hazard by also increasing the expected size of the resulting cornice fall avalanche. Furthermore, in all cornice fall avalanches observed on Gruvefjellet the main cornice blocks traveled further downslope than the rest of the avalanche debris. This pattern is apparent on larger failures on Platåberget as well, but is in some cases less obvious, likely due to the smaller cornice blocks being functionally indistinguishable from the avalanche debris. While the dataset presented here is insufficient to draw more quantitative conclusions regarding the runout distance of these cornice blocks, hazard management strategies should consider the destructive potential and extended runout of these blocks relative to the other entrained snow.
The TLS data acquisition and processing techniques employed in this work allowed us to illustrate and quantify changes to the observed cornice systems in detail not previously achieved, but our results and subsequent interpretations are nonetheless limited by several factors. Measurement uncertainties specifically related to measuring snow surfaces with TLS are well-discussed in previous research (Deems et al., 2015; Prokop, 2008), but we introduced additional uncertainty to our results and interpretations due to the scan timing. Our TLS data acquisition scheme involved time-intensive manual input, so we were unable to achieve the temporal resolution required to better constrain individual accretion and cornice failure events. Decreasing time between scans would allow for more continuous and robust accretion rate calculations and could better constrain failure and avalanche snow surfaces, especially pre-event. Sufficiently decreasing the between-scan interval to a sub-daily resolution for such applications would likely require some degree of automation, and future work should consider employing a permanently installed TLS acquiring data automatically similar to systems employed for mining applications or slope stability assessments.
Uncertainties in cornice volume calculations are also affected by occasionally lengthy inter-scan intervals. Volume changes corresponding to specific meteorological conditions are in these cases aggregated across the entire scan interval, making disentangling the specific contributions to volume changes difficult. These conceptual uncertainties are magnified by the technical uncertainties related to TLS data acquisition. The TLS accuracy is of increased importance for volume quantification as measurement uncertainties are propagated throughout the volume calculation process. However, calculated volume uncertainties (Appendix A) are sufficiently low to instill a degree of confidence in the volume calculation process presented here. Finally, volume calculations are perhaps least robust in this study for times when the lack of obvious cornice structure makes calculating volumes particularly challenging (e.g., Platåberget during the 2017–2018 season).
Our experimental design focused on investigating the evolution and failure of the lower cornice surfaces from scan positions underneath the cornices where accessibility has precluded previous research. These scan positions did not, however, allow for systematic monitoring of the cornice roof. The orientation of the cornices' leading edges frequently shielded the cornice roof from the scanner, and our profiles often do not include the complete cornice roof. This also has implications for representative volume calculations, as uncertainty in the location of the cornice roof can result in inaccurate horizontal difference calculations in these specific locations. By failing to capture the cornice roof in our data, we also limit comparisons with earlier work on Gruvefjellet relating downslope cornice deformation and cornice failure to the appearance of tension cracks between the cornice roof and the plateau anchoring point (Vogel et al., 2012). Future work should pair TLS data with some form of tension crack observation, and approaches combining TLS and unmanned aerial vehicle (UAV) photogrammetry present intriguing possibilities for future work in this and other locations.
TLS was shown to be a particularly suitable remote sensing tool for cornice monitoring in Svalbard where we were able to obtain useful data during the early winter seasons when the polar night precludes direct visual observation and cornice photography. Svalbard's unique environmental characteristics – such as the polar night – limit to a degree the applicability of our results to lower latitudes where more diurnal variations in radiation and temperature may influence cornice dynamics in ways not represented in Svalbard (e.g., Munroe, 2018). It is also unclear how representative the two winter seasons for which we present data are for the cornice systems in Longyeardalen, as previous research has also noted considerable differences in cornice dynamics between seasons (Vogel et al., 2012). Continued cornice monitoring in this and other lower-latitude settings would help clarify such uncertainties.
We monitored seasonal cornice dynamics and associated cornice fall avalanche
activity with a TLS over two winter seasons in high-Arctic Svalbard. The
spatial and temporal resolution at which we acquired snow surface data with
the TLS allowed us to quantify changes to the cornices with sub-decimeter
accuracy. These data provide quantitative reinforcement to existing conceptual models of cornice dynamics and further strengthen the validity of
these models. Notable quantitative contributions from this work include
documentation of conservatively calculated horizontal accretion rates well
in excess of 10 mm h
This study demonstrated the viability of TLS methods for monitoring cornice dynamics. TLS methods for obtaining snow surface data are appropriate in Svalbard where the long polar night precludes data acquisition via other methods (e.g., photogrammetry), but techniques presented in this work are also suitable for cornices in other lower-latitude environments. Future work should investigate automated TLS data acquisition as an avenue to improve the temporal resolution of the measurements and better constrain cornice dynamics to specific meteorological conditions.
Our findings show complex interactions between topography, wind speed and direction, snow available for transport, existing snowpack, and cornice structure govern the growth, failure, and associated avalanche activity of the cornices in Longyeardalen. In particular, we show cornices rapidly accrete given winds strong enough to mobilize surface snow from a direction roughly perpendicular to the plateau edge, placing the cornices in the lee. Our findings also reinforce previous work indicating an increased likelihood of cornice failure and associated avalanche activity during these periods of cornice accretion. This is encouraging for hazard managers seeking to forecast cornice fall avalanches, as anticipating the relatively infrequent conditions leading to cornice accretion can help predict periods of elevated cornice fall avalanche hazard. We observed the largest failures in our dataset in areas with minimal topographic support, demonstrating knowledge of the topography underlying the cornices can be beneficial when considering the specific location of cornice failure. Nevertheless, our limited dataset of cornice failures hinders conclusions drawn from this work, and continued work in a variety of environments is needed to better understand the specific mechanisms and dynamics of cornice fall avalanches.
TLS data summary for the 2016–2017 winter season. n/a denotes “not applicable”.
n/a
TLS data summary for the 2017–2018 winter season.
Data from the Svalbard Airport AWS are freely available through the Norwegian Meteorological Institute's web portal, eKlima (eKlima, 2019). Data from the Gruvefjellet AWS are freely available through the University Centre in Svalbard (UNIS) via
HH was responsible for the majority of the data acquisition, analyses, and interpretation of the results. ME helped develop the conceptual framework for the study and contextualized and interpreted the results within a broader snow and avalanche perspective. AP provided technical guidance with regards to TLS data acquisition and analysis techniques and assisted in the development of the study's technical framework in addition to assisting in data acquisition. JH provided advice and supervision relating to study design, data analysis, and interpretation of the results. HH and ME were responsible for manuscript preparation with input from AP and JH.
The authors declare that they have no conflict of interest.
We thank Christine Fey and Jeffrey Munroe for their thorough and constructive reviews which greatly improved this work. Andreas Günther is thanked for serving as editor for this work.
This paper was edited by Andreas Günther and reviewed by Christine Fey and Jeffrey Munroe.