Evolution of events before and after the 17 June 2017 landslide at Karrat, West Greenland – a multidisciplinary approach for studying landslides in a remote arctic area

The 17 June 2017 rock avalanche on the south facing slope of the Ummiammakku Mountain (Karrat Isfjord, West Greenland) caused a tsunami that flooded the nearby village of Nuugaatsiaq, killed four persons and destroyed 11 buildings. Landslide activity in the area was not previously known and the disaster gave rise to important questions about what events led up to the landslide and what the future hazard is in the area around the landslide? However, the remoteness 15 of the area and difficult fieldwork conditions, made it challenging to answer these questions. We apply a multidisciplinary workflow to reconstruct a timeline of events on the coastal slope here collectively termed the Karrat Landslide Complex. The workflow combines limited fieldwork with analyses of freely available remote sensed data comprising seismological records, Sentinel-1 space borne Synthetic Aperture Radar (SAR) data and Landsat and Sentinel-2 multispectral optical satellite imagery. 20 Our analyses show that at least three historic rock avalanches occurred in the Karrat Landslide Complex: Karrat 2009, Karrat 2016 and Karrat 2017. The last is the source of the tsunami and the first two are described for the first time here. All three are interpreted to have initiated as translational rockslides. In addition to the historical rock avalanches, several pre historic rock avalanche deposits are observed, demonstrating older periods of activity. Furthermore, three larger areas of continuous activity are described and may pose a potential future hazard. A number of non-tectonic seismic events 25 confined to the landslide complex are interpreted to record landslide activity. Based on the temporal distribution of events in the landslide complex, we speculate that the possible trigger for landslides is permafrost degradation caused by climate warming. The results of the present work highlight the benefits of a multidisciplinary approach based on freely available data to studying landslides in remote Arctic areas under difficult logistical field conditions and demonstrates the importance of 30 identifying minor precursor events to identify areas of future hazard. https://doi.org/10.5194/esurf-2020-32 Preprint. Discussion started: 11 May 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
On 17 June 2017 the village of Nuugaatsiaq in West Greenland was hit by a tsunami generated by a 35-58 million m 3 landslide on the south facing slope of the Ummiammakku mountain in Karrat Isfjord located 32 km to the east of the village (Bessette-Kirton et al., 2017;Clinton et al., 2017;Gauthier et al., 2018;Paris et al., 2019). A large part of the 35 village was destroyed, and four people lost their lives. The tsunami was also observed in other settlements more than 100 km away. Following this, the Greenlandic authorities evacuated 170 residents from Nuugaatsiaq and the neighbouring settlement of Illorsuit due to the threat of further landslides in the area and the villages continue to be evacuated at the time of this writing due to fear of additional landslide triggered tsunamis (Fig. 1). The Karrat Landslide highlights the necessity to screen the inhabited parts of Greenland for unstable slopes and to map previous large landslides to assess the 40 risk of future tsunamigenic events.

Fieldwork and in situ measurements are difficult and time consuming in a vast and remote Arctic environment like
Greenland where infrastructure is minimal and expensive. Thus, investigations of unstable slopes over large parts of Greenland must primarily rely on remote sensing techniques. Following the launch of Landsat-8 and Sentinel-1 and 2 satellites, optical and SAR data over Greenland is both free and frequent and at sufficient resolution, providing a means 45 of observing deforming slopes at relatively low cost. Svennevig et al. (2019) developed a multi-disciplinary approach combining satellite SAR, optical data, and seismic observations to study remotely activity on an unstable slope. They found that by combining these methods it was possible to reliably detect timing (seismic observations), location, extent and deformational rates (optical images, deformation from DInSAR (Differential Interferometric Synthetic Aperture Radar)) of landslide activity.

50
Our aims with this study are twofold: to understand the processes that led to the catastrophic Karrat 2017 rock avalanche, and to evaluate the risk of further landslides at Ummiammakku Mountain. Following the multi-disciplinary approach of Svennevig et al. (2019) it is possible to resolve the series of events leading up to and following the disaster in Karrat Fjord in June 2017. We show that it would not be possible to establish both timing and location of all events based on one method alone. We contextualize our results using geological knowledge of the area derived from limited fieldwork and 55 previous studies and discuss the possible landslide trigger mechanism.
landslides also occurred. The geologic province where the 2017 landslide occurred was found to have relatively few landslides (Fig. 1a). Field observations show that the Palaeoproterozoic rocks on the slope parts easily along distinct layering of the bedding (s0 foliation) which dips 20 to 30° to the south towards the fjord. Furthermore, E-W orientated

Methods and data
We use a workflow integrating seismological data, SAR and optical imagery -all publicly available -for describing the evolution of the Karrat Landslide Complex. These data sources have different temporal and spatial resolution ranging 75 from years to milliseconds and meters to 10's of kilometres (see table 1). Individually they have unique information for studying landslides, but tell an incomplete story by themselves and the value of the individual datasets increases significantly when integrated. The workflow is described and applied in Svennevig et al. (2019) examining a minor (ML 1.9) non tectonic seismic event in the Karrat Landslide Complex on 26 March 2018.
We found that alerting each other across disciplines of suspected smaller landslide events enabled us to construct a reliable 80 multi-year sequence of both confirmed smaller landslides and periods of activity in the area. For example, if a seismic event was suspected of being caused by a landslide, optical satellite images before and after the time of the seismic event was inspected for changes, and InSAR (Interferometric Synthetic Aperture Radar) images constructed for evidence of movement. Alternatively, if optical satellite images showed change between two satellite passages, we could check if a seismic event had occurred in the area in the time interval, and if InSAR analyses showed movement to confirm either 85 minor activity or indeed a landslide.

Fieldwork
Because of the remoteness of the area and the steepness of the coastal slope carrying out fieldwork is logistically challenging. Because of the continued risk of landslides (see below) and near constant minor rockfalls (Fig. 2E) it is not 90 safe to come closer than about 1.5 km of the landslide area. These conditions highlight the need for remotely sensed data as exemplified below. We visited areas just east and west of the Karrat 2017 rock avalanche on two short reconnaissance stops during the summer of 2019 to make observations of the surrounding geology and to inspect the landslide area using a camera-equipped multirotor UAV (Unmanned Aerial Vehicle).

95
DInSAR Slope deformation can be detected remotely using DInSAR-based techniques (Carlà et al., 2019;Rosen et al., 2000), which provide one-dimensional ground-motion measurements between two radar acquisitions in a radar equipped satellite line-of-sight direction, i.e. towards and away from the radar. The Sentinel-1A and -1B are Earth monitoring synthetic aperture radar (SAR) satellites operated by the European Space Agency (ESA) that covers all of Earths landmass. The 100 data produced from these platforms is freely available distributed by the ESA. The Sentinel-1 data acquired over land provide 5 m x 20 m spatial resolution in the ground range and azimuth (flight-path) directions respectively. Two satellite https://doi.org/10.5194/esurf-2020-32 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License. tracks cover the Karrat area during the period of interest: ascending track 90 (available from October 2014) and descending track 25 (available from July 2017). The viewing geometry of track 25 is best suited for detecting movements on the slopes in our region of interest. Unfortunately, large parts of the steep slopes that failed in 2016 and 2017 cannot 105 be observed with the viewing geometry of ascending track 90 that covers the pre-failure time period. SAR data is insensitive to cloud cover and the polar night as opposed to optical satellite data from e.g. Sentinel-2 (see below).
However, a prerequisite for the applicability of DInSAR is a sufficient level of statistical similarity (interferometric coherence) between the electromagnetic properties of the surface at the two acquisition times. This can be lost due to changes in the satellite viewing geometry or physical changes at the surface between acquisitions. Ground motion of more 110 than half a wavelength (2.8 cm for Sentinel-1) between acquisitions will cause a complete loss of coherence, called decorrelation, in the image. In practise, decorrelation occurs at lover ground motion due to the other factors affecting the coherenceSentinel-1 data are acquired every 6 days over Greenland, and for this study both 6 and 12 days differential interferograms were used. The topographic contribution to the interferometric phase was removed using ArcticDEM

Seismology
The Geological Survey of Denmark and Greenland (GEUS) monitors seismic activity in Greenland using the Greenland 120 Ice Sheet Monitoring Network (GLISN network, www.glisn.info), which consists of 21 stations (Clinton et al., 2014).
Detecting and accurately locating the activity in the Karrat area depends on having a sufficient number of nearby stations (see Fig. 1A). The stations in this area are located along the coast with a distance of at least 100 km between them. Thus, the horizontal location uncertainty of detected earthquakes or other types of seismic events is up to 50 km, in particular in the east-west direction.

125
Not only tectonic earthquakes are detected in Greenland. We see many events that we classify as non-tectonic events.
This class of events were first described by Ekström et al. (2003) and were found to be located at Greenland's large outlet glaciers. The monitoring carried out by GEUS locate many non-tectonic events smaller than the globally detected events described by (Ekström et al. (2003), and also many of these are located close to large outlet glaciers. The causes of nontectonic events are several. For example, events with epicentre located near an outlet glacier (cryo-seismic events) often 130 contain a low frequency component, and are usually much longer in duration than tectonic earthquakes (Fig. 5), and are interpreted to be caused by calving of glaciers (Ekström et al., 2003;Nettles et al., 2008). Other non-tectonic events, in Western Greenland, are mainly caused by sea ice breakup, glacier or sea ice movements on bedrock, but other types are also present see e.g. Podolskiy and Walter (2016). Suddenly failing landslides also produce a seismic signal. The Karrat 2017 rock avalanche was seen globally as a Ms 4.2 event (U.S. Geological Survey, 2020), and the 2000 Paatuut landslide 135 was seen throughout Greenland (Dahl-Jensen et al., 2004). Smaller events associated with known landslides (this paper) are only seen more locally (Fig. 5).
Non-tectonic events can easily be identified from tectonic earthquakes based on their different frequency content and P and S amplitudes (Fig. 5).
However, distinguishing a landslide signal from other non-tectonic events, such as events associated with glaciers, is not For smaller landslides the tremor component will be smaller in duration and amplitude. Many aspects of smaller known landslide events are similar to cryo-seismic events (Fig. 5). There are several large outlet glaciers in the Karrat area responsible for non-tectonic events within the horizontal uncertainty of the location of events, so the location in itself is 145 not sufficient to distinguish whether the source is a glacier outlet or possibly a landslide. In the Discussion section of this paper a first step towards distinguishing between these two types on non-tectonic events is described.
Seismological data enabling us to register and locate smaller non-tectonic events became available around 2008. The first station in Greenland was in operation 1906-1912. In the late 1920's two more permanent stations were installed. Until  https://doi.org/10.5194/esurf-2020-32 Preprint. Discussion started: 11 May 2020 c Author(s) 2020. CC BY 4.0 License.

175
The Landsat program is a series of Earth monitoring multispectral optical satellites, the first of which, was launched in 1972. Landsat 1-5 (1972Landsat 1-5 ( -1993 had spatial resolutions of 60 m and Landsat 6-8, a 30 m resolution. Landsat-7 and -8 revisits the same area every eight day since the launch of Ladsat-8 in 2013. Further back in time the coverage is sparser.
As is the case for Sentinel-2 scenes Landsat images at this high latitude have a winter data gap from mid-October to start-March. Data are freely available from USGS within 24 hours of acquisition.

Aerial images
To constrain the evolution of the Karrat Landslide Complex a set of 1:45 000 scale black and white aerial images from 1953 (available from The Danish Agency for Data Supply and Efficiency) have been analysed. These constitutes the oldest known aerial images from the area.
185 Table 1: Temporal and spatial resolution of the various datasets.

Results -landslides and active areas at the Karrat Landslide Complex
In order to describe the multifaceted landslide evolution of the Karrat area it is necessary to establish a nomenclature framework. Hence, we introduce the Karrat Landslide Complex as a 3 by 9 km area of past, present and future landslide   Optical data is not very helpful in this case due to darkness, although it is in agreement with InSAR. Analysis of the 225 seismic signal reveals a magnitude ML 2.1 non tectonic event to take place at 15 November 2016 at 11:34Z (Fig. 5D).
The westernmost part of the slide scarp is visible in an ArcticDEM strip from 5 June 2017. Based on this DEM and the geometric constraints of the scarp of the Karrat 2009 rock avalanche we calculate the volume to be 3.0 x 10 6 m 3 . It is not possible to constrain the volume of the deposit from the landslide as no DEM covers the entire area. However, based on the Sentinel-2 images, it seems that some of the material may have reached the fjord (Fig 3C). Bessette-Kirton et al. Area 2 (71°38'46"N, 52°21'57"W) Area 2 is a 500 by 700 m well developed slump located 500 m west of the Karrat 2017 rock avalanche at 950-1200 m elevation (Fig. 1B, 2C). The area could not be visited during fieldwork due to the steepness of the terrain and the near constant rock falls (Fig. 2E). Drone inspection of the 50 m high back scarp showed that bedrock is exposed there (  (Fig. 5f). The event could not be confirmed by InSAR and optical interpretation due to poor data coverage in the short period between the event and the later Karrat 2017 rock avalanche. It is thus reported here as a seismic event that is possibly a landslide.
The events up until the time of submission are listed in This study shows the effectiveness of combining complementary remote sensing techniques to establish precise time and location of a long series of landslide activity. It is an inexpensive setup relying on freely available and continuously updated datasets. Although much can be accomplished without fieldwork, the multidisciplinary approach cannot stand alone: It is an effective tool for identifying and investigating active landslide areas, but actual field validation is necessary in order to further assess the risk.

Evolution of the Karrat Landslide Complex
As our compilation of results from the multi-disciplinary approach show, the Karrat 2017 rock avalanche was not an isolated event, but part of an ongoing process of landslide erosion focused in the Karrat Landslide Complex, (Fig 1A and   Fig. 6B). This erosion can be subdivided into one or several earlier phases and a series of recent events since 2009. Area 2 starts to be active from May 2015 as seen in InSAR and a seismic event (Fig. 4A). In addition, localised  Table 2). The absence of seismic events and landslides in the decades before the Karrat 2009 rock avalanche is taken as an indication that the area was relative dormant prior to this time (Fig. 3A).

350
The east to west migration of the rock avalanches in the eastern part of the Karrat Landslide Complex suggests a westward migration of a fracture acting as back scarps for the three historical landslides. This, along with the ongoing deformation detected by InSAR in Area 2 and 3 (Fig. 4d) and 2017 rock avalanches seems to be situated just below sea level to the west below Area 1 and might act as the basal sliding surface of this also (Fig. 1B).

Possible trigger mechanisms
It has not been possible to determine exactly what triggers individual landslides and seismic events in the Karrat area and 360 why there seems to be a recent peak (since 2009). However, the events are distributed throughout all seasons and, from the limited data we have available, no seasonal change in activity can be seen (Fig. 6C). This indicates that something with longer period than the seasonal cycle could be at work.
We hypothesize that the slope instability is induced by permafrost degradation (Draebing et al., 2014;Krautblatter et al., 365 2013). The regional air temperature has increased by 4-5 °C since 1880 and has been accelerating since c. 1990 (Cappelen et al., 2018). The prehistoric activity in the complex described here could have taken place during a previous climatic optimums such as the Holocene optimum where climatic conditions in the arctic are thought to be similar to those of today (Axford et al., 2019). The subsequent slow cooling could have stabilized the slopes again until the recent warming.
With the projected temperature increase of up to 8 °C towards 2100 (IPCC, 2013) a range of landslide risk factors is 370 expected to increase, including permafrost degradation and thusly the risk of landslides from the Karrat Landslide Complex could be expected to increase.
A variety of methods could be applied to test this hypothesis such as dating pre-historic (Holocene) landslide activity, analysing aerial images from the past century to constrain historic evolution (creep) and installing climate sensors to 375 constrain the present permafrost conditions of the slope. Bathymetrical studies of the seabed just of the Karrat Landslide Complex could also be included.

Regional hazard evaluation and context
As a whole, landslide intensity in the Karrat area (geological area of Proterozoic metasediments interfolded with Archean 380 gneiss: Fig. 1) is not particularly higher than elsewhere in Greenland (Svennevig, 2019). In this context, the Karrat Landslide Complex is a local entity and an outlier with respect to landslide intensity as neighbouring slopes in the fjord system with similar types of bedrock show no abnormal landslide activity. This indicates that local conditions on the slope are responsible for the high intensity such as s0 parallel dipslope weaknesses and subvertical fracture systems (Fig.   2F) along with possible permafrost degradation. The regional landslide hazard in the Karrat area, with the exception of 385 the Karrat Landslide Complex, is thus not considered to be higher than elsewhere in Greenland.
A consequence of the multistage evolution of the Karrat Landslide Complex is that Gauthier et al. (2018) and Paris et al. also suspected cryogenic seismic events that are of the same duration as suspected landslide events. Currently, we must rely on supporting evidence from the satellite data in order to confirm of dismiss a suspected landslide events seen seismically.
A denser local seismograph network in central West Greenland has been rolled out during the summer of 2019. This will improve the location accuracy of events in the area -including the Karrat Landslide Complex -allowing event location 420 to help separate non-tectonic events into cryogenic seismic events and potential landslide events.

Conclusions and outlook
This study shows the effectiveness of using the multi-disciplinary setup described in Svennevig et al. (2019) for studying landslides in remote Arctic areas. We show that the disastrous Karrat 2017 rock avalanche was not a single event. Smaller previous studies (Gauthier et al., 2018;Paris et al., 2019) have overestimated the volume of the catastrophic avalanche.
The volume of material entering the fjord is an important input to tsunami models and will have implications for estimating the range of the generated wave and ultimately for the risk assessment related to future landslides from the areas with continued activity.
The consequence of a tsunami from a "worst case" landslide from the Karrat Landslide Complex to the two abandoned