Introduction

The Assapaat frozen debris avalanche occurred in mid-June 2021 on the south coast of Nuussuaq (70°19ʹ09ʺN, 52°59ʹ48ʺW) facing the Vaigat strait, Central West Greenland (Fig. 1). Nearby large landslides have previously failed disastrously in December 1952 and November 2000 (Pedersen et al. 2002; Dahl-Jensen et al. 2004). Both historic events triggered large waves that we will in the following call ‘tsunami’ for simplicity. Although, strictly speaking this term comes from the Japanese and means ‘harbour wave’ which does not fit here. Several other terms are in use such as landslide-triggered tsunami, displacement wave, non-seismic wave, giant wave, and impact wave (see also discussion in Hermanns et al. 2013). However, the Geological Survey of Denmark and Greenland (GEUS) used the term ‘tsunami’ in earlier publications (e.g. Pedersen et al. 2002; Dahl-Jensen et al. 2004; Clinton et al. 2017; Svennevig et al. 2020) and we will use it for consistency. The term is also internationally widely used to describe high water waves caused by landslides impacting into a water body (Sepúlveda and Serey 2009; Harbitz et al. 2014; Higman et al. 2018). The historic tsunamis in Vaigat thus demonstrate the potential threat posed by landslides in this area. The 1952 event resulted in one fatality and material damage in the town of Qullissat, 30 km from the source, and the 2000 event resulted in severe material damage to Qullissat, 20 km from the source, and minor damages to the village of Saqqaq 50 km away. For this reason, the slopes of the Vaigat strait were recognized as an area with a high risk of tsunamigenic landslides (Pedersen et al. 2002; Dahl-Jensen et al. 2004). Similarly, in 2017, a large rock avalanche near Karrat, 150 km to the north, caused a tsunami that resulted in four fatalities in the village of Nuugaatsiaq, 30 km away. This incidence resulted in the evacuation of Nuugaatsiaq and another nearby village Illorsuit (Paris et al. 2019; Svennevig et al. 2020). Due to currently active slopes in the vicinity of the 2017 rock avalanche, the two villages are still evacuated for the fifth year to the date of writing (Svennevig et al. 2020).

Fig. 1
figure 1

A Historical landslides in Vaigat (red polygons) shown on a simplified 1:100,000 scale geological map. Also shown are the two local seismic stations at Paatuut (red triangles) and the position (red star) and uncertainty ellipsis (red stippled line) of the seismic event mentioned in the text. Contour interval is 100 m. The insert map shows the position of A in Greenland and the six regional seismic stations (red triangles) where the event was recorded. Readings from the PATT, GDH, and SAATT stations are shown in Fig. 2B. The yellow star denotes the position of the Karrat 2017 rock avalanche and the villages of Nuugaatsiaq and Illorsuit mentioned in the text. B False colour Sentinel-2 image (bands 8, 4, and 3) from Sentinel Hub EO Browser from 18 June 2021. Vegetation is shown in red highlighting bare cliffs and the area of the frozen debris avalanche (dashed magenta outline) as dark grey. Contour interval is 50 m. Positions of photos and maps in Figs. 3, 5, and 7A are shown

Full size image
Fig. 2
figure 2

Seismology. A Seismic time series of the stations PATT, SAATT, and GDH in the time of the detected event related with the Assapaat frozen debris avalanche. All three stations feature three components (Z, N, E), and the data is filtered with a zero-phase bandpass filter applying corner frequencies of 1–6 Hz. The PATT station was situated at Paatuut and equipped with a 4.5-Hz geophone period sensor. SAATT and GDH are broadband stations within the operational regional network in West Greenland. Picked phase arrivals used for the localization of the seismic event are indicated by a solid orange line (P wave) and a dashed pink line (S wave). In total, the PATT and six stations of the regional network were utilized for the event localization. B Blow-up of the seismic event connected to the Assapaat frozen debris avalanche. The data is from the vertical component (DPZ) of the PATT station and shown as seismogram and spectrogram. To highlight its low frequency content of about 3 Hz, the data was resampled with 100 samples per second

Full size image
Fig. 3
figure 3

Pre-failure photos of the scar area. A Oblique photo from 27 August 2018, showing the source area 3 years prior to the event. The outline of the Assapaat frozen debris avalanche is shown by a dashed purple line, and the dominant InSAR anomaly (see Fig. 4) is shown by the dashed yellow line. An active scarp on the lower slope is labelled with ‘S’, and old vegetated debris cones (molards) are labelled with ‘M’. An active deformation fracture in the colluvium below the talus slope of the source area is labelled with ‘F’. A seasonal stream draining from the plateau into the top of the talus slope of the source area is labelled with ‘W’. The top of the clouds in the bottom of the left photo is at approximately 300 m elevation. B Oblique photo from 19 August 2019. Small cracks (‘C’) developed in the talus slope. Dashed outlining and labelling are the same as in A

Full size image

The landslide record of polar regions is incomplete due to the remoteness and sparsely populated nature of these areas. However, in recent years a number of studies have been published: Eisbacher (1979) reported on Holocene rock avalanches in north western Canada, presumably associated to tectonic activity. Kuhn et al. (2019, 2021a, b) reported failures of reactivated rock slides from Svalbard. Coe et al. (2018) reported an increase in rock avalanche size and mobility during the last 32 years for the Glacier Bay National Park Alaska. Furthermore, previously unreported historical rock avalanches have been lately registered in northernmost Norway (Hermanns et al. 2021b). Generally, it is important to expand the Arctic landslide record by analysing recent and historical events and so fill data gaps of rock slope deformation and failure. A more complete landslide record will help to better address hazard assessments of potentially enhanced failure of Arctic rock slopes in a warming climate (Patton et al. 2019; Coe 2020; Sæmundsson et al. 2021).

We combine analysis of pre- and post-failure remote sensing data (Sentinel-1 and -2), seismic records, and morphological mapping in the field, along with analysis of a pre- and post-failure digital elevation model (DEM), to give a detailed description of the 2021 Assapaat frozen debris avalanche. We describe the large amount of entrainment that occurred during that event that we relate in the discussion to permafrost degradation. We furthermore propose a hypothesis for the large failure of the talus slope that initiated the frozen debris avalanche, and associate this to warming of the permafrost. We finish this contribution with a brief discussion about the impact of a warming climate on the hazard from this type of landslide in the Arctic.

Study area and geological setting

The study area is located in central West Greenland on the south coast of the Nuussuaq Peninsula facing the Vaigat strait (Sullorsuaq in Greenlandic) (Fig. 1A). The topography is shaped by repeated Quaternary glaciations with up to 2000-m-high coastal slopes along the 600-m-deep, 15-km-wide, and 100-km-long northwest to southeast striking Vaigat strait. The strait is located 60 to 150 km west of the Inland ice and was deglaciated between 12 and 10 ka BP (Weidick and Bennike 2007). The present climate in central West Greenland is Arctic and the ground in the region is continuously permafrozen (Westergaard-Nielsen et al. 2018); however, the permafrost conditions on steep slopes are virtually unknown. According to DMI, the Danish Meteorological Institute (www.dmi.dk), the mean annual air temperature is − 8.4 °C at sea level in the settlement of Saqqaq, 50 km southeast of Assapaat. At present, West Greenland is tectonically stable with only minor earthquakes probably associated with plate tectonics or isostatic rebound (Voss et al. 2007).

The bedrock geology of the coastal slopes of Vaigat consists of planer bedded Cretaceous-Palaeocene sediments of the Nuussuaq Group (Dam et al. 2009) exposed up to 800 m elevation, overlain by extensive Palaeogene basalt and volcanic breccia (hyaloclastite), and intruded by associated sills and dykes. The volcanic and subvolcanic rocks are associated to the opening of the Labrador Sea and the continental drift of Greenland over proto-Icelandic mantle plume (Chalmers and Laursen 1995; Henriksen et al. 2009). Locally at Assapaat, poorly lithified mudstones and sandstones of the Albian-Paleocene Nuussuaq Group form the bedrock up to 600 m elevation. These are overlain by Paleocene hyaloclastite breccias and a local subaqueous volcanic eruption site of the Vaigat Formation (Pedersen et al. 2017). A topographic ledge is developed at the contact between the 15°–30° sedimentary slope and the 60°–80° 200-m-high hyaloclastite cliff. This ledge is filled with colluvium and till capped by a thick active talus slope from weathering and spalling of the hyaloclastite cliffs. As the till and colluvium appear to be intermixed, this deposit is likely predating the deglaciation 12–10 ka BP.

Previous catastrophic rock slope failures in Western Greenland

The detailed circumstances of the 1952 Niiortuut landslide have not been described except for a comment in the paper on the 2000 Paatuut rock avalanche (Pedersen et al. 2002). Work on the 2000 Paatuut rock avalanche by Pedersen et al. (2002) and Dahl-Jensen et al. (2004) proposes that the avalanche mainly involved in situ bedrock of Palaeogene volcanic rocks sliding on unconsolidated Palaeogene mudstones. They suggested that this was the common model for landslides in this geological region—the Nuussuaq Basin. The rock avalanche consisted of 90 * 106 m3 of material, of which 60 * 106 m3 was deposited in the subaerial part of the lobe on the lower slope. 30 * 106 m3 entered the strait causing a tsunami with a run-up height of 50 m on the nearby coast.

In a preliminary inventory of post-glacial landslides of Greenland, the Nuussuaq Basin was identified as an area with high density of landslides, occupying just 4.7% of the ice-free area of Greenland, but featuring 67% of the mapped landslides (Svennevig 2019). The recent 2017 Karrat rock avalanche occurred 150 km north of the Assapaat landslide, outside the Nuussuaq Basin (Svennevig et al. 2020). The bedrock in this geological province consists of Proterozoic metasediments of the Karrat Supergroup and Archean basement (Henriksen et al. 2009). Svennevig et al. (2020) concluded that the failure was conditioned by the local dip slope and speculated that permafrost degradation was a preparatory factor.

Morino et al. (2019) describe remotely sensed decametre scale conical mounds termed molards on an old landslide deposit 4 km west of Assapaat. Such mounds, especially when they undergo thaw collapse, are indicative of permafrost degradation Morino et al. (2019), although care must be taken in such interpretations because landslide mounds can also form in non-permafrost regions (Paguican et al. 2014; Blais-Stevens et al. 2015; Dufresne and Geertsema 2020). Similar landforms have been observed in the deposit of the 2000 Paatuut rock avalanche (Pedersen et al. 2002), and we have now observed a high concentration of these in the recent Assapaat landslide deposit, which will be further addressed in the “Results” and “Discussion” section.

Methods and data

Our analysis is based on morphological mapping of the Assapaat frozen debris avalanche in the field and remote sensing data as well as seismological data from a regional network and two local stations.

Oblique photos, digital elevation model, and 3D modelling

Pre-event oblique photos of the coastal slope were acquired during routine helicopter flybys on 27 August 2018 and 6 August 2019 and are used here to show the conditions of the slope before the event. Overlapping post-event photos were collected from a helicopter on 31 July and 17 August 2021, using a Nikon D850 digital SLR camera equipped with a Carl Zeiss Distagon 1.4/35 mm lens and a Solmeta GMAX geotagging GPS unit. A subset of the images recorded in July 2021 was used to construct a post-event DEM with a spatial resolution of 1 m following the workflow outlined in Sørensen and Dueholm (2018). Briefly summarized, the DEM was produced in a two-step process. Firstly, the orientation of the images was established by solving a bundle block adjustment within the AnchorLab software ‘3D Stereo Blend’ using a combination of GPS data for the images, tie-points, sea-level levelling points, and ground control points obtained from a pair of 1:40,000 pre-event aero-triangulated vertical photographs acquired in 2008. Secondly, the DEM was produced using the NFrames software SURE and with the calculated image orientations as input data. Subsequently, a 1-m-resolution pre-event DEM was produced from the aero-triangulated vertical photograph control source using the NFrames software SURE. Finally, a DEM of difference (DoD) was produced by subtracting the post-event DEM from the pre-event DEM to quantify the net sediment budget. To estimate the total material involved in the event, the base of the landslide was modelled in the 3D modelling software MOVE constrained by the post-event DEM, the DoD, and the field observations.

Seismology

Seismic activity in Greenland is operationally monitored by a regional network of seismic broadband stations (Clinton et al. 2014). Since 2010, the network in Central West Greenland is dense enough to register and locate non-tectonic events such as rock avalanches and smaller rock slides (Dahl-Jensen et al. 2004; Svennevig et al. 2019, 2020). At present, the regional seismic stations are mainly located along the west coast of Greenland with an interstation distance of at least 100 km (Fig. 1). Thus, the location uncertainty of detected earthquakes, or other types of seismic events, is up to 50 km. Two seismic stations were installed from August 2020 to August 2021 to monitor rockfall activity at Paatuut, 10 km east of Assapaat, one three-component station, and one mini-array station, both equipped with very short period sensors (Fig. 1A). We have manually screened all available seismic data from the regional network and the two stations at Paatuut for a seismic event potentially caused by the frozen debris avalanche within the possible period of the event (11th to 14th of June).

Interferometric Synthetic Aperture Radar (InSAR)

Slope deformation can be detected remotely by spaceborne InSAR (Colesanti and Wasowski 2006; Lauknes et al. 2010; Eriksen et al. 2017). The conventional differential InSAR (DInSAR) method allows for qualitative distinction between stable and unstable areas in each analysed time interval. For this study, DInSAR was applied to imagery from the Sentinel-1 twin satellites operated by ESA, the European Space Agency. We analysed 59 images from the descending track 25 acquired between June 2018 and October 2021. The descending satellite line-of-sight direction of track 25 is well suited to detect the southwest motion of the unstable slopes along the south coast of Nuussuaq. Unfortunately, only one of the Sentinel-1 satellites has been acquiring data in the area since 2019, so the minimum time span for differential interferograms is 12 days. The interferometric coherence of image pairs is usually limited by snow cover from November to May. Differential interferograms were formed for image pairs with enough interferometric coherence using the SARPROZ software (Perissin and Wang 2011), applying a 4 × 1 averaging (multilooking), resulting in ground resolution cells of approximately 20 m × 20 m. The topographic contribution to the interferometric phase was removed using ArcticDEM (Porter et al. 2018). Under favourable conditions, the Persistent Scatterer (PS) method may monitor slope deformation with millimetre precision. In this study, the PS method was applied to images acquired prior to the Assapaat frozen debris avalanche. However, the deformation rate of the unstable slope in the source area is too high to allow for identification of PS with the available number of images.

Thermal imaging

Thermal images of the scarp area were acquired from a DJI Mavic 2 Enterprise multirotor unmanned aerial vehicle (UAV), equipped with a dual visible light/IR camera. The images were taken 50 days after the event to constrain the relative temperature of the back scarp and document degrading permafrozen blocks.

Results

Based on preliminary observations such as a long runout, a carapace of large boulders of frozen colluvium that decays into debris cones (molards (Milana 2016; Morino et al. 2019), see below), and a large landslide volume, we classify the landslide as a frozen debris avalanche (sensu Hungr et al. (2014)). We name it the ‘Assapaat frozen debris avalanche’ after Assapaat, the boulder field present on the coastal slope before the event. We will discuss the landslide type in more detail in the “Discussion” section after presenting our morphological and volumetrical analyses.

The event itself was not observed by eyewitnesses. The area was first observed by a local resident sailing by on 14 June 2021 at 19:08 local time who took a short video sequence of the area (see Data Repository 1). Initial screening of Sentinel-2 images did not reveal the exact day of the event due to cloud cover. However, the last cloud free image showing an undisturbed slope is from 9 June, and the first usable one after the event from 18 June, thus confirming the locality of the frozen debris avalanche and bracketing the timing (see Data Repository 2). Local sources describe the slope as still unaffected by the event on 11 June at 17:32 local time. This limits the timing of the frozen debris avalanche to the period between the evening of 11 June and the evening of 14 June. No tsunami initiated by the frozen debris avalanche was reported by locals and field observations and the available satellite images likewise showed no signs of anomalous wave activity (Fig. 1B).

A seismic event was recorded on 13 June, at 7:04 (4:04 local time), located a few kilometres from the frozen debris avalanche, well within the uncertainty ellipse (Fig. 1A). The event was registered on the local Paatuut stations (PATT, PATW), as well as on six stations in the regional network (SAATT, GDH, UPNV, SUMG, NEEM, ATTUG) with a magnitude of ML 1.0 (Fig. 2). The event duration is about 20 s. The PATT station’s P and S wave arrivals exhibit a time difference of 2.3 s (Fig. 2B) which corresponds to a distance of 25 km applying a Greenland-wide 1D velocity model (Stevens et al. 1972). The distance derived from the seismic data is in the same order of magnitude as the actual distance between the Assapaat frozen debris avalanche and the PATT seismic station (~ 10 km). The distance difference can be explained by the available velocity model which is too coarse to reflect local and shallow velocity variations in the Vaigat region.

Pre-failure situation

Oblique photos from August 2018 and August 2019 show the upper part of the slope with boulders and large debris cones covered by arctic vegetation indicating an older (Early Holocene(?), tentatively based on the degraded morphology) episode of landslide activity at the site (Fig. 3). A seasonal stream from the plateau above the coastal slope is draining into the top of the talus slope. The source area of the 2021 frozen debris avalanche consists mainly of colluvium dominating the active talus slopes. The lower part of the colluvium, below the talus slope, is presumably of mixed origin consisting of older landslide deposits and till. A large fractured (sheared) body of hyaloclastite bedrock is seen protruding through the upper part of the talus slope in the source area. Differential InSAR images outline an unstable area in the upper talus slope, which corresponds well to the later source area (Fig. 4). Prior to 2020, the interferograms show a coherent deformation pattern corresponding to approximately 1 m/year in line-of-sight. In 2020, the creeping velocity of the slope increased, as shown by the increasing decorrelation in the 12-day interferograms, and the unstable area was expanding towards west (Fig. 4). The 2021 interferograms prior to failure could not be used as they are generally noisy due to atmospheric disturbance and snow cover. Despite the observed deformation in the InSAR data, no indicative morphological features such as the development of a clear back scarp, or flanks, are visible in the 2018 photo. An exception is a 180 by 150 m area of minor fractures in the colluvium below the talus slope in the south-eastern part of the source area (‘F’ in Fig. 3). An oblique photo of the same site in August 2019 shows the situation to be roughly similar although small cracks developed in the talus slope (‘C’ in Fig. 3B). In the middle part of the slope, at around 300 m elevation, a 100-m-wide and 15-m-high scarp is visible (‘S’ in Fig. 3A), corresponding to an observed InSAR anomaly (‘S’ in Fig. 4).

Fig. 4
figure 4

Differential 12-day interferograms from descending track 25 for July 2018 and August 2020. The interferogram is overlain by a pre-event transparent hillshade and contours from the ArcticDEM. The outline of the Assapaat frozen debris avalanche is shown with a black dashed line, and unstable areas affected by the frozen debris avalanche are marked with white polygons. Several other areas in the vicinity of the Assapaat frozen debris avalanche also show movement. Note the slight widening towards west of the unstable source area from 2018 to 2020. The widening and increasing decorrelation of the source area indicates accelerating slope deformation. S points to the same area as S in Fig. 3A

Full size image

Morphological features of the Assapaat frozen debris avalanche

The primary source area is defined by the primary scarp and coincides with an InSAR anomaly seen in interferograms from July 2018 and August 2020 (Fig. 4). The uppermost part of the primary scarp runs along the face of the 60° steep cliff of hyaloclastite bedrock at 700 to 880 m elevation. The western part of the primary scarp cuts through the talus slope and underlying colluvium down to 600 m elevation. There is no indication that the hyaloclastite bedrock was involved in the frozen debris avalanche (Fig. 6A, B). The lower part of the source area, defined by the basal failure surface at 700 to 600 m elevation, is covered by debris from the event itself and is not exposed, however can be roughly delineated by the lowermost exposure of the back scarp (Figs. 5, 6).

Fig. 5
figure 5

Morphological map. Main morphological components of the Assapaat frozen debris avalanche. Post-event hillshade and contours are shown as the background. Positions and directions of photos in Fig. 6 are indicated in green. See Fig. 1B for location within the Vaigat coast

Full size image
Fig. 6
figure 6figure 6

Field photos. See Fig. 5 for photo locations and directions. A Oblique aerial image of the Assapaat frozen debris avalanche from 9 August 2021. B The primary source area recorded on 17 August 2021, showing the perched talus slope along the hyaloclastite cliff. The slope at the bottom part of the photo is 500 m across. Positions of the melting permafrozen blocks in G and H are shown on the right side of the photo. Persons for scale (encircled) are standing on the top of the western part of the scarp. C The western part of the secondary scarp at 400 m elevation. The mud below the large boulder is covering a surface of massive interstitial ice. D Levée at 200 m elevation formed by local soil and debris cones (molards). E Debris cones and pressure ridges at the foot of the western lobe of the frozen debris avalanche. The western lobe is 200 m wide. F Secondary mudflow activity in the eastern lobe at 100 m elevation. G Thermal image (purple and black indicate relatively cooler surface temperatures) and H optical image of the same scene inside the primary scarp recorded on 31 July 2021 showing melting blocks of permafrost-cemented talus. The dark colouring in H is caused by water from the melting interstitial ice. Also notice the cold/dark colouring of the main scarp in the background in the upper left corner of photos G and H. I Detail of the permafrozen dark part of the central block shown in G and H. The pore space is filled with massive ice covered here by a thin veneer of mud

Full size image

Midway down the slope, from 400 to 250 m elevation, lateral scarps into the coastal slope, up to 20 m high, are developed along the western and eastern margin of the frozen debris avalanche defining a 700-m-wide secondary scarp (Figs. 5, 6A). At the western flank, this coincides with the scarp that was visible on the August 2018 photo (‘S’ label in Figs. 3A and 4A). At the base of the western scarp, a mud-covered surface of massive ice is observed, which we assume to be the basal failure surface of the secondary scarp (Fig. 6C). This surface lies at a relative depth of 20–30 m below the neighbouring undisturbed surface of the slope. The lower part of the basal sliding surface of the secondary scarp below 380 m elevation is not exposed. It is covered by a 1050-m-long and 650-m-wide convex mass of debris cones and boulders of hyaloclastite comprising the main frozen debris avalanche deposit (Figs. 5, 6A). Along the western margin of the main lobe, below the secondary scarp, an up to 10-m-high levee composed mainly of locally derived soil and debris cones was developed in places (Figs. 5, 6D).

From about 130 to 100 m elevation, the deposit splits into three 190–460-m-wide and up to 500-m-long lobes, stretching to the coast and propagating 125–190 m into the Vaigat strait (Figs. 5, 6A, E). These lobes comprise mainly debris cones of colluvium composition and no fresh hyaloclastite boulders were found here. The lobes propagating into the sea feature concentric ridges. In the western part of the deposit, two smaller, up to 50 m wide and 200 m long, lobes terminated 30 to 75 m above the coastline. The 460-m-wide eastern lobe is eroded by a gully originating from c. 370 m elevation (Figs. 5, 6A). Mud within this gully suggests that a secondary mudflow contributed to this gully erosion (Figs. 5, 6F). Analysis of Sentinel-2 images shows that this occurred between the 18th and the 19th of June where a small delta built out east of the eastern lobe and that the delta continued to build out up to the 24th of June (see Data Repository 2).

Throughout the whole deposit, colluvium debris cones up to 20 m in diameter are observed (Fig. 6A, B, E). Hardly any loose debris was seen in the field. During the fieldwork 45–60 days after the event, several of the debris cones showed water escape structures from their flanks. We suggest thawing of interstitial ice as the source of water. Probing a conical mound at an elevation of ~ 300 m with a field hammer c. 60 days after the event indicated that the interior of the mound was frozen from a depth of 10 cm. In the primary scarp at an elevation of 650 m, several blocks of permafrost-cemented talus with massive interstitial ice in the matrix and c. 20 m in diameter were observed. At the time of observation, c. 45 days after the event, the blocks were in the process of thawing and degrading into debris cones (Fig. 6G, H, I). We saw no obvious signs of liquid water being involved in the frozen debris avalanche. Water escape structures from the debris cones and the mudflow activity in the eastern gully were both clearly secondary events.

The top of the primary scarp is at 880 m elevation and the lowest visible part of the deposit is at sea level, 2480 m horizontal distance from the top scarp. The submerged extension of the deposit is unknown, as no nearshore bathymetric data is available. The linear interpolation between the pre-event shoreline and the nearest bathymetrical data delivers a slope angle of 7° for the nearshore seabed. Assuming an angle of repose of 30° for the submarine part of the lobe, the deposit would continue at least 30 m vertically and 50 m horizontally from the shoreline. This results in a minimum total vertical drop of 910 m, and a minimum total horizontal distance of 2530 m for the Assapaat frozen debris avalanche, thus in a H/L ratio of < 0.36 corresponding to a fahrböschung (angle of reach) of < 19.8.

DEM of difference (DoD) and volume reconstruction

The DoD delivers a lower bound for the total sediment budget of the frozen debris avalanche because our field and DoD observations that revealed significant frozen debris avalanche deposits inside the primary and secondary scarp (Figs. 6A, B, 7A). Thus, we cannot find the total volume of the frozen debris avalanche from the DoD alone. In the following, we first present the results of the DoD analysis in the form of net volumes. Then, we estimate the total (gross-) volume of material involved inside the primary and secondary scarp. We do this by modelling the base of the frozen debris avalanche within the constraints of the DoD and field observations. An overview of the sediment budget and modelled volumes is given in Table 1.

Fig. 7
figure 7

A DEM of difference (DoD, June 2008–31 July 2021) showing the change in terrain elevation for the Assapaat frozen debris avalanche. The red and green stippled lines delimit the identified zones of net erosion and deposition identified in the field and DEM analysis. The volumes from the DoD calculations are shown in their respective zones. The outline of the frozen debris avalanche is shown with a purple stippled line. B Topographic profile through X-Xʹ in A, showing the topography before (red) and after (black with light yellow fill) the event. The bathymetry is interpolated. The black dash-dotted lines are the modelled basal failure surfaces covered by deposits. (I) represents the modelled minimum extent of the secondary scarp, and (II) the maximum extent. Grey triangles are the minimum extent of the frozen debris avalanche in the profile. C Vertically exaggerated DoD profile along the X-Xʹ profile. Net erosion is shown in red and net deposition is shown in green. The identified erosional and depositional zones are shown by the labelled curly brackets. The colour of the bracket labels indicates erosion (red), or deposition (green). Interpreted basal failure surfaces are shown as dashed-dotted black lines

Full size image

Based on the DoD, the eroded area can be divided into three subzones (Fig. 7; Table 1): (1) the primary scarp exposed from 880 to 600 m elevation, where the frozen debris avalanche was initiated and 3.9 * 106 m3 of talus slope and colluvium was removed; (2) a 260-m-wide erosional channel from 600 to 450 m elevation, down the 35° slope where 2.4 106 m3 was removed; and (3) a 650-m-wide and up to 20-m-deep secondary scarp developed from 450 m down to at least 300 m elevation, where the slope angle is 15° and where 3.3 * 106 m3 was removed.

Table 1 Volume budget from DoD and 3D modelling
Full size table

Deposition can also be divided into three zones: (1) deposition inside and just southwest of the primary scarp at 600 m elevation. Just southwest of the primary scarp a 0.1 * 106 m3 lobe is present; (2) deposition of the main lobe from 450 m down to 100 m elevation, partly inside the secondary scarp, where a net volume of 4.8 * 106 m3 is deposited; (3) deposition in three minor lobes from 100 m elevation to sea level, and volumes of 2.1 * 106 m3, 0.5 * 106 m3, and 0.6 * 106 m3, respectively. A bulking factor of 1.25 was applied to the net erosion of 9.6 * 106 m3 to account for a volume increase of 25% due to fragmentation during transportation. This is similar to the factor suggested by Hungr and Evans (2004) and commonly used for rock avalanches (e.g. Schleier et al. 2015; Oppikofer et al. 2017). The sediment budget is thus a net mobilisation of 12.0 * 106 m3 minus the net deposition of 8.1 * 106 m3. The difference of 3.9 * 106 m3 must have been deposited distal to the three minor lobes in the Vaigat strait (Table 1).

The basal sliding surface of the primary scarp was reconstructed by two planes: The upper part was approximated by a plane parallel to the exposed basal sliding surface. This is rotated 30° relative to the profile plane and dipping 30°. The lower part was modelled as a plane perpendicular to the profile plane, dipping 5° and intersecting the terrain just above 600 m (Fig. 7B, C). The volume between this surface and the pre-event topography is 6.9 * 106 m3 (Table 1), which is taken as the maximum volume of the talus slope that failed.

As only very little deposition is observed inside the channel, the volume from the DoD of 2.4 * 106 m3 is assumed as the total eroded volume of the channel.

Only the upper part of the basal sliding surface of the secondary scarp is exposed, so a maximum and minimum possible volume is modelled and constrained by observations from the field and the DoD. The upper part of the sliding surface is modelled as a 16.5° dipping plane perpendicular to the profile plane from 450 m elevation. The lower part is modelled as a 5° dipping plane perpendicular to the profile plane intersecting the terrain at 280 m as the minimum, and at 110 m as the maximum estimate (Fig. 7B, C). This is based on the field and DoD observations that this surface was not exposed above or below these elevations. Intersecting these surfaces with the pre-event topography gives a modelled volume for the slope inside the secondary scarp ranging from 8.4 * 106 to 14.5 * 106 m3.

The modelled volumes of the scarps and the determined volume of the channel are multiplied by 1.25 to account for a 25% increase in volume due to fragmentation during transport (bulking factor, Hungr and Evans (2004)). This gives gross volumes of 4.8 * 106 m3 for the initial frozen debris avalanche originating at the primary scarp (accounting for the volume still inside the primary scarp) and entraining 3.0 * 106 m3 in the channel (VFʹ). The combined volume of 7.8 * 106 m3 impacted the slope at 450 m elevation and further entrained between 10.5 * 106 and 18.1 * 106 m3 of material in the secondary scarp. The total volume estimate of the frozen debris avalanche is thus between 18.3 * 106 and 25.9 * 106 m3.

To examine the amount of entrainment, an entrainment ratio (ER; sensu Hungr and Evans 2004) is calculated. This is not straightforward due to the multiple stages of entrainment. The initial frozen debris avalanche first deposited 3.1 * 106 m3 in or just outside the primary scarp. A volume of 5.5 * 106 m3 was evacuated from the primary scarp and entrained 3.0 * 106 m3 of the channel, and further the 10.5 * 106 to 18.1 * 106 m3 of the secondary scarp. We therefore first calculate the ER for the channel entrainment (ERʹ), and thereafter for the secondary scarp as a range (ERʹʹ). ERʹʹ is calculated for both the maximum and minimum modelled volume entrained by the sum of the volume of the initial frozen debris avalanche plus the volume entrained from the channel (VF + VEʹ = 8.5 * 106 m3). Finally, we calculate a total entrainment ratio ERtotal for the entrainment of both the channel and the secondary scarp (VEʹ + VEʹʹ) by the initial frozen debris avalanche (VF).

The ERʹ for the channel by the initial frozen debris avalanche is thus 0.6 (3.0 * 106 m3 / 5.5 * 106 m3). ERʹʹ for the secondary scarp is between 1.2 and 2.1 (10.5 * 106 m3 / 8.5 * 106 m3 and 18.1 * 106 m3 / 8.5 * 106 m3) and ERtotal is between 2.5 and 3.8 (13.5 * 106 m3 / 5.5 * 106 m3 and 21.1 * 106 m3 / 5.5 * 106 m3).

Discussion

Remote sensing data and field observations allowed us to reconstruct the June 2021 frozen debris avalanche. However, the event was not observed and essential data for determining the preconditioning of the event are not available (e.g. local air and ground temperature, water saturation of the soil, snowmelt, and precipitation). Such data are in general missing for slope collapses that were not predicted, and for remote arctic events, such as the Assapaat frozen debris avalanche, regional climate and weather data are also in short supply. Conditioning factors can therefore only be deduced and discussed based on observations following the event.

We have in the “Introduction” established that the landslide is a frozen debris avalanche. Further classifying this landslide is, however, not straightforward for two reasons: (1) The material involved was mainly debris (talus, colluvium) which would make the landslide a debris avalanche (sensu Hungr et al. 2014). However, it is important to stress that the debris was cemented by permafrost and behaved mechanically as a rock avalanche: a granular flow of boulder material (blocky material, very to extremely rapid movement (Hungr et al. 2014; Hermanns et al. 2021a) (see below), before impacting the lower slope. (2) After the frozen debris avalanche impacted the lower slope, it entrained the material inside the secondary scarp and transitioned into a relatively slower dry debris avalanche and widened downward in the colluvial deposits (sensu Hungr et al. 2014). Thus, we define the entire landslide as a ‘frozen debris avalanche triggering a dry debris avalanche’. For simplicity, we shall keep referring to the landslide in its entirety as ‘frozen debris avalanche’.

Seismic event

The seismic signal we interpret as related to the frozen debris avalanche event was small (ML 1.0; Fig. 2), and it took extended analysis to detect. Since the event is at the detection limit of the regional network, it would probably not have been found without the two local Paatuut stations, only 10 km away from Assapaat (Fig. 1). This is in contrast to previous observations of rock slope failures in the area that produced unambiguous seismic signals detected by the regional network and featured comparable source-station distances (Dahl-Jensen et al. 2004; Svennevig et al. 2019, 2020). In the Karrat area, 150 km north of the Assapaat frozen debris avalanche, two precursor rock avalanches occurred in September 2009 (2.8 * 106 m3) and November 2016 (3.0 * 106 m3), prior to the giant (41–43.5 * 106 m3, ML 4.2) June 2017 rock avalanche (Svennevig et al. 2020). The volumes of the precursors were an order of magnitude smaller than the Assapaat frozen debris avalanche, but nevertheless produced seismic signals of magnitude ML 2.7 and ML 2.1, respectively, clearly observed on the regional seismic network. Our strongest indication for the seismic event being related to the frozen debris avalanche is the localization at Assapaat within the time period of interest. The event waveform features a single P and S phase, and a rather short event duration (Fig. 2B). Both features point towards an instantaneous source impulse, rather suggesting a local tectonic event than a rock slope failure. The classic seismic signal of a rock avalanche is stimulated by the superposition of multiple wave type phases caused by the multiple impacts of tumbling down boulders. This results in an emergent seismic signature often described as a seismic coda wave with no clear phases of distinct wave types (Dammeier et al. 2011; Hibert et al. 2015). However, Dahl-Jensen et al. (2004) reported a seismic event with single P and S phases also for the close-by Paatuut rock avalanche in November 2000. Furthermore, the frequency content of the favoured event is around 3 Hz (Fig. 2B). A local tectonic event this close to the station would feature higher frequencies, hence supporting the interpretation that the event is rather related to the frozen debris avalanche than to a local tectonic event. Eventually, due to the present uncertainties, we refrain from any process-based interpretation of the seismic signal and use it just for a more accurate timing of the Assapaat frozen debris avalanche. Furthermore, we want to stress the surprisingly low seismic energy generated by the relatively large Assapaat frozen debris avalanche, which made the event detection and localization a tedious undertaking with the nonetheless remaining uncertainties.

The June 2021 Assapaat frozen debris avalanche development

The Assapaat frozen debris avalanche initiated with a failure of 6.9 * 106 m3 of permafrost-cemented colluvium of mainly talus on the coastal slope of Vaigat. The failure depth was about 40 m below the original slope surface inside the permafrozen slope and at an elevation of 600 to 880 m. A volume of 5.5 * 106 m3 (after bulking factor) of the failed material continued out of the primary scarp as a frozen debris avalanche. Moving down the 30° slope, the frozen debris avalanche entrained 3.0 * 106 m3 of colluvium and carved a 260-m-wide and 23-m-deep channel. The erosive nature of the channel and narrow elongate shape lead us to infer that the movement of the frozen debris avalanche was very to extremely rapid here (sensu Hungr et al. 2014). At 400 m elevation, the slope flattens to an angle of ~ 15°. Here the now 8.5 * 106 m3 large frozen debris avalanche impacted and initiated a large secondary failure. This subsequent failure entrained between 10.5 * 106 and 18.1 * 106 m3 of permafrozen colluvium down to a depth of 20–30 m (Figs. 6C, 7). The depth of the active layer on the lower slope is unknown, but we assume that it is substantially shallower than 20–30 m, and that the secondary failure happened inside the permafrost body. Active layer thickness of 2.8 m has been observed in 2021 at a borehole at 1000 m elevation c. 60 km South East of Assapaat. The light dry/warm top layer seen in Fig. 6G and H in the melting permafrozen block at 600 m elevation is 1–1.5 m thick and may be taken as a proxy for the active layer in the source area. Pre-failure InSAR interferograms and photos show that the colluvium inside the future secondary scarp was stable in the years up to the event, except for a minor area in the western part (Fig. 4). We thus interpret the primary failure, the channel erosion, and the secondary scarp failure as three sequential cascading landslide events initiated by the failure in the primary scarp. This is based on the geomorphological observations that the primary scarp, channel, and the secondary scarp are all constrained by separate erosional morphologies and on the observation that the secondary scarp eroded into the undeformed slope and with a clearly defined back scarp (Figs. 5, 6A, C).

The total volume of the frozen debris avalanche is 18.3 * 106 to 25.9 * 106 m3, and most of this volume is deposited inside or just below the secondary scarp. Below the secondary scarp, the main deposit is flanked by levees (Figs. 5, 6D), indicating that the energy was lower in this part of the landslide relative to higher on the slope.

Topographically below the main lobe, three subordinate lobes continued down the slope and about 3.9 * 106 m3 of material (Table 1) entered the sea corresponding to between 15 to 20% of the total volume. The central and western lobes display concentric ridges (Figs. 5, 6A, E) that we interpret to be pressure ridges due to momentum loss when entering the sea resulting in relatively lower velocities in this part of the frozen debris avalanche. The absence of evidence of a tsunami along the coast despite the submergence of a substantial volume supports the suggestion that the landslide featured a velocity below extremely rapid in its most distal part.

Reported rock avalanches with similar volumes have triggered several tens of meter high tsunamis in the near-field of the impact zone (Roberts et al. 2014). Previous landslides in Vaigat in 1952 and 2000 caused destructive tsunamis (Pedersen et al. 2002; Dahl-Jensen et al. 2004). The circumstances of the 1952 landslide are largely unknown but a reconnaissance visit during the recent fieldwork showed the deposit to be dominated by molards and thus also potentially be of the frozen debris avalanche type, but this landslide warrants further detailed investigation. The 2000 Paatuut rock avalanche was channelized. The rock avalanche, thus, probably had enough speed and energy to produce a destructive tsunami, although a submarine failure of unknown size was suggested to also have played a role in the development of the tsunami (Pedersen et al. 2002; Dahl-Jensen et al. 2004). Our recent findings regarding the Assapaat frozen debris avalanche warrant a reexamination of the 2000 Paatuut rock avalanche.

Conical hummocks are found throughout the deposit (Fig. 6). We interpret these as degraded (thawed) permafrozen blocks of talus termed molards (Milana 2016; Morino et al. 2019). This is supported by three observations: (1) c. 45 days after the event, frozen blocks of talus in the process of thawing were observed inside the primary scarp (Fig. 6G, H). (2) Water escape structures from melting of the interstitial ice within the molards were observed throughout the frozen debris avalanche deposit. (3) Digging into a molard at ~ 300 m elevation showed that the interior was still frozen. The abundance of molards in the deposit implies that the majority of the involved material was permafrozen when it failed, and consequently, behaved mechanically as boulders of ice-cemented talus during transport (see Morino et al. 2019). Furthermore, we interpret the vegetated debris cones observed on the slope prior to the frozen debris avalanche (M in Fig. 3A) as ancient molards, indicative of previous, ancient slope failures of frozen colluvium in post-glacial times. This is also supported by the large, sheared body of hyaloclastite bedrock inside the talus slope, along with the numerous boulders in between the molards.

Entrainment on the permafrozen slope

As summarized above, two distinct phases of entrainment can be inferred along the path of the frozen debris avalanche based on morphological evidence: (1) within the channel on the 30° steep slope, and (2) inside the secondary scarp where the slope angle flattens to 15°. The circumstance that no tsunami was observed suggests that the overall speed of the frozen debris avalanche was below extremely fast while entering the strait. Consequently, and based on morphological evidence (i.e. the presence of a levee, concentric pressure ridges in the distal part (Figs. 5, 6D, E)), we infer that the secondary phase of entrainment inside and below the secondary scarp decelerated the frozen debris avalanche. This is probably due to transfer of energy from the smaller initial frozen debris avalanche to the larger volume within the secondary scarp. The flattening of the slope gradient from 30° to 15° could also have caused the downslope kinetic energy of the initial frozen debris avalanche to be directed into the slope and thus be a contribution factor to the entrainment. At this position on the slope, the entrainment ratio increases from 0.6 to between 1.2 and 2.1. A decrease in flow velocity as we, based on morphological interpretation, observe is in contrast to the common observation that entrainment of substrate material often increases mobility of rock avalanches (Hungr and Evans 2004; McDougall et al. 2006; McDougall 2017). Geertsema et al. (2006) describe a case from British Colombia, Canada, similar to the Assapaat frozen debris avalanche, where a decrease in velocity is observed following entrainment. In the Assapaat case, the increase of the entrainment ratio occurs in the same section as the decrease of the slope angle. Thus, the amount of material that had to be accelerated from the stable slope within the secondary scarp (entrained material) to the mean frozen debris avalanche velocity increased at the same location as deceleration due to topographic conditions (slope gradient) occurred. Pudasaini and Krautblatter (2021) describe this in a recent paper as related to momentum transfer and erosional velocity of the landslide. The latter is dropping sharply below the secondary scarp at the Assapaat frozen debris avalanche as shown by our volume calculation. Also, Chen et al. (2020) showed through numerical modelling of the 2017 Xinmo landslide that entrainment of dry material on the slope leads to more frictional energy consumption and a reduction of landslide mobility. We therefore conclude that entrainment of the large dry permafrozen volume inside the secondary scarp together with the low slope gradient consumed a large fraction of the kinetic energy of the frozen debris avalanche.

Latent heat for melting of interstitial ice could have acted as another sink for (1) the energy released by the impact of the smaller initial frozen debris avalanche, and (2) the frictional heat generated by the downslope travelling material. Rapid melting of ice by internal friction and impact energy during the Kolka/Karmadon rock-ice avalanche has been hypothesised (Huggel et al. 2005). The abundance of molards and absence of mudflows or other evidence of large amounts of water during the frozen debris avalanche indicate that the amount of ice melted during the event itself must have been limited. Any amount of thawed material amongst which the blocks of frozen debris travelled downslope would have resulted in viscous and frictional dissipation of energy and contribute to explain the observed low seismic energy.

Conditioning and triggering factors for the initial frozen debris avalanche

Conditioning factors for the initial frozen debris avalanche can be divided into static conditioning factors: those that do not change in time, such as lithology and the structural setting, and dynamic conditioning factors; and those that change the stability of the slope (preparing factors) and control the timing (triggering factors) of rock slope failures (Hermanns et al. 2006).

The static conditioning factors of the Vaigat coast are favouring rock slope failures as shown by both the historic (Pedersen et al. 2002; Dahl-Jensen et al. 2004) and the Holocene landslide record (Svennevig 2019). The static conditioning factors for the 90 × 106 m3 Paatuut rock avalanche in 2000 were suggested to be unfavourable bedrock geology where competent layers of Paleocene volcanic rocks failed along underlying Paleocene unconsolidated and less competent mudstones (Dahl-Jensen et al. 2004). Similar conditioning settings have been suggested for numerous Holocene landslides present within the area (Dahl-Jensen et al. 2004). This contrasts with our observations from the Assapaat frozen debris avalanche where the initial failure happened in the talus slope and the bedrock was not involved. The top of the scarp developed along the bedrock-talus interface and penetrated 40 m into the permafrozen talus slope, and further propagated inside the colluvium below the talus. This indicates that in case of the Assapaat frozen debris avalanche, the static conditioning factors are rather related to the talus slope and colluvium, than to the hyaloclastite and sandstone-mudstone bedrock. An important factor may be that the talus slopes are over-steepened. However, the talus slope that failed in the frozen debris avalanche is 30–32° (Fig. 7B)—as are the neighbouring talus slopes. This is within the normal range for talus slopes (Chandler 1973; Carson 1977) and implies that a steepness above the friction angle of granular material is probably not the conditioning factor.

On longer time scales than seasonal, progressive warming of the permafrozen slopes toward a stability threshold is a likely dynamic conditioning factor and has already been speculated for other Arctic sites (Svennevig et al. 2020; Kuhn et al. 2021b). Limited literature exists on how deformation in permafrozen talus slopes initiates even though this topic crosses into the vast rock glacier literature (see e.g. Janke and Bolch 2021). In the following, we propose a hypothesis for this and the static and dynamic conditioning factors of the talus slope failure at Assapaat (Fig. 8). We base our interpretations on our field observations as well as laboratory tests by Davies et al. (2001) and Krautblatter et al. (2013), who demonstrate that ice-filled fractures become less competent as temperature approaches the melting point and have the lowest shear strength at c. − 0.5 °C and that hydrostatic pressure can generate larger destabilising shear stress down to significant depths into cracks with connectivity to surface hydrology. Warming and the infiltration of water to the depth of a shear horizon similarly cause higher rates of deformation in ice-cemented debris, as documented for rock glaciers (Kenner et al. 2020), which are a widespread landform in this part of Greenland (Humlum 1996, 2000). The above leads us to the following hypothesis for the evolution of the slope failure:

  1. 1.

    As the talus slope builds up by rock fragments spalling from the cliff above, the clasts have a grain-grain contact and the friction between the clasts stabilises the slope at the typical angle of repose of granular material of 30–32° ((1) in Fig. 8A). Some ice may already accumulate at this stage from occasional burial of snow and may form massive ice lenses. Water from precipitation and snowmelt percolates into the talus slope and porosity completely fills with interstitial ice below the active layer similar to rock glaciers’ accumulation of interstitial ice (Janke and Bolch 2021). The talus slope is sufficiently ice-cemented to have very low hydraulic permeability. This configuration is stable, with no or very slow creep due to the interstitial ice being significantly colder than the melting point, which we assume was the case for the previous centuries with colder than present mean air temperatures (Vinther et al. 2006).

  2. 2.

    With mean air temperature increasing compared to previous centuries (Vinther et al. 2006), the permafrost will start to warm. This causes slope deformation to accelerate markedly, potentially due to the activation of shear planes along horizons of more massive ice whose shear strength has dropped. This faster deformation results in open fractures especially at the backscarp that allow surface waters from snow melt and precipitation to penetrate behind the still-impermeable bulk of the ice-cemented talus, raising the hydrostatic pressure and further lowering the shear strength until the steep slope angle of 30–32° exceeds its stability point.

Fig. 8
figure 8

Failing talus slope hypothesis. A Static conditioning factors: the slope configuration. The talus slope is building up on the older till and colluvium by spalling from the cliff and the clasts feature a grain-grain contact except where lenses of massive ice from buried snow bodies exist. The friction between the grains stabilises the slope at an angle of 30–32°. If the internal slope temperature is significantly colder than the melting point (Davies et al. 2001), the interstitial ice is strong enough to balance the occurring shear stresses inside the ice lenses. B Dynamic conditioning (preparatory) factors: warming of the slope body. As the air temperature increases, the slope body warms from the surface inwards. When the ice matrix approaches the melting point, the occurring shear stress of the slope body exceeds the shear resistance of the interstitial ice and ice lenses. Consequently, the top slope body starts to creep downwards, and open fractures develop along the cliff face. This makes it possible for water (blue wave) to penetrate deep into the slope and act as a trigger

Full size image

The assumptions in this hypothesis (past and present temperature within the slope, the nature of the interior of the talus slope, etc.) merit further investigation in view of their potential relevance for other similar slopes, which is however beyond the scope of this paper.

Movement rates of at least 1 m/year were observed in InSAR interferograms of the talus slope in the years prior to the failure (Fig. 4). Based on field evidence of the depth to the surface of rupture of 40 m and consistently with the conceptual model described above, we assume that the observed surface movement rates are mainly constrained to the top 40 m of the frozen talus slope (Fig. 7C). Thus, preparing factors must have worked at this time interval and depth (Fig. 8B). Oblique field images from 2018 and 2019 indicate that the development of the talus slope from incipient movement to failure potentially happened relatively fast (Fig. 3) as also indicated by the increasing area of decorrelation seen in InSAR between 2018 and 2020. Because the talus slopes and underlying colluvium were permafrozen prior to and during failure, warming of the permafrost body by higher air temperatures is a good candidate for a preparing conditioning factor working on these time scales and depth in the slope.

The trigger of the frozen debris avalanche is likely the product of a suite of dynamic conditioning factors (Hermanns et al. 2006). We found no apparent external trigger for the initial frozen debris avalanche as no earthquakes or exceptional precipitations—snowmelt or temperature episodes—have been reported prior to the event. However, weather observations from the greater area are rather scarce. Dahl-Jensen et al. (2004) interpreted the trigger of the Paatuut rock avalanche to be weather conditions (freeze–thaw-freeze) in the days prior to the slide. The Assapaat frozen debris avalanche occurred in mid-June 2021, the 1952 Niiortuut landslide in mid-December, and the 2000 Paatuut rock avalanche in mid-November (Pedersen et al. 2002; Dahl-Jensen et al. 2004). Other recent rock avalanches at Karrat, 150 km to the north, occurred in September 2009, November 2016, and June 2017 (Svennevig et al. 2020). Although this is a small sample and the individual triggers are unknown, we can conclude that there is so far no clear seasonal pattern in the timing of larger rock slope failures in Central West Greenland. This is unsurprising, at least with respect to annual air temperature as a trigger, when considering that the failure surfaces are deeper than the 10–15 m commonly assumed to be the maximum depth of zero annual amplitude of temperature (French 2017).

Accelerating deformation from the permafrozen slope gradually warming over years can promote the opening of deep reaching fractures at the top of the talus slope, behind the still impermeable ice-cemented talus body further down the slope (Fig. 8B). Even in the absence of heavy precipitation of snowmelt events, infiltration of surface waters over time would produce a perched aquifer and the associated hydraulic loading (Fischer et al., 2010). Thus, preconditioned failure may be triggered at any time once the increased hydrostatic pressure at the backscarp and at shear horizons at depth exceeds their shear resistance Krautblatter et al. (2013).

We observed seasonal snowmelt in Sentinel-2 and InSAR data in the weeks up to the frozen debris avalanche and observed a seasonal stream draining into the top of the talus slope during past and present snowmelt seasons (Fig. 3; Data Repository 1). However, the seemingly random distribution throughout the year of the other historic rock slope failure events mentioned above indicates that this trigger would only count for failures that have occurred in the late spring to early autumn season. A failure surface at a depth of 40 m as observed would support that hydrostatic pressure in the back scarp could have triggered the slide. Further investigation into the other historic (and future!) events in the region could help to constrain this better, but it is unlikely that enough data on slope conditions and weather exists.

Climate change and long-term hazard development perspective

Annual mean air temperature in the Arctic is projected to increase by 7.5 °C by 2100 (RCP 4.5) relative to the mean temperature from 1900 to 1950, and already today the Arctic mean air temperature increased by 2 °C since preindustrial time (Overland et al. 2019). Recent rock slope failures along the Vaigat coast and the Karrat Fjord in Central West Greenland have been documented, but there are currently insufficient data to show a clear correlation between these and the recent Arctic warming. Increased rock avalanche activities in the past decades have been observed in several more well-studied Arctic, Sub-Arctic, and alpine areas with permafrost conditions (McSaveney 2002; Huggel 2009; Fischer et al. 2010; Guthrie et al. 2012; Coe et al. 2018; Bessette-Kirton et al. 2020), and have also been observed in places associated to earlier Holocene warm intervals (e.g. Hilger et al. 2018). Furthermore, accelerated movement and reactivation have been reported for permafrost-controlled unstable slopes in Norway and on Svalbard (Blikra and Christiansen 2014; Böhme et al. 2019; Etzelmüller et al. 2021; Hilger et al. 2021; Kuhn et al. 2021b).

We have suggested that the Assapaat frozen debris avalanche, and the large-scale entrainment of the permafrozen slope it triggered, is a result of permafrost warming, causing higher rates of deformation and leading to the formation of open fractures reaching shear horizons at depth. That this is due to the ongoing climatic warming is thus an obvious assumption to make. As such, the Assapaat frozen debris avalanche is a fingerpost towards the long-term hazard development of Greenland and the Arctic in general. Permafrozen slopes can fail catastrophically, potentially activating and entraining large volumes of warming permafrozen sediments, and in some historic cases trigger destructive tsunamis. This natural hazard scenario is important for large coastal talus slopes. A further aspect is that the volume of the initially unstable slope was much smaller than that of the eventual total frozen debris avalanche. Generally, the entrainment potential of landslides is likely to further increase in the future as permafrozen slopes continue to warm, and eventually thaw (Etzelmüller et al. 2021). Due to the local topography, the Assapaat frozen debris avalanche was slowed down by the entrainment of the volume inside the secondary scarp. However, other colluvium covered slopes in the Arctic are much steeper. In the currently warming climate, it is likely that future rock slope and frozen colluvium failures meet thawed and water saturated colluvium along their run-out path which might in contrast have an accelerating effect on landslide mobility. This would consequently increase the threat of catastrophic tsunamis which was not observed for the Assapaat frozen debris avalanche.

Conclusions

The June 2021 frozen debris avalanche at Assapaat, on the south coast of Nuussuaq in central West Greenland, started as a 6.9 * 106 m3 failure of permafrozen colluvium of mainly talus slope at 600–880 m elevation. The frozen debris avalanche entrained a large volume of permafrozen colluvium along its path in two subsequent entrainment phases resulting in a total volume between 18.3 * 106 and 25.9 * 106 m3. The second phase of entrainment and a flattening of slope angle reduced the mobility of the frozen debris avalanche. Even though 3.9 * 106 m3 of material entered the Vaigat strait, no tsunami was evident.

The Assapaat frozen debris avalanche generated surprisingly low seismic energy which made the detection and localization of the related seismic event a tedious undertaking, still leaving the discussed uncertainties. Enhanced local seismic monitoring of sensitive regions like Vaigat poses a feasible strategy to overcome these remaining uncertainties for comparable future events.

Our analysis of the event leads us to the hypothesis that the talus slope failure was preconditioned by warming of the ice-cemented talus slope. The deformation rate made it possible for opening fractures to let water infiltrate the otherwise impermeable permafrozen slope, elevating hydrostatic stress and lowering shear resistance to the failure point.

We attribute the initial frozen debris avalanche and the large-scale entrainment of permafrozen material to historic warming of the slopes and suggest that continued warming of the Arctic might lead to an increase in the frequency and magnitude of rapid landslides in the Arctic region. This will result in an increased threat of landslides and their secondary and cascading consequences.