Paraglacial exposure and collapse of glacial sediment: The 2013 landslide onto Svínafellsjökull, southeast Iceland

In the last 130 years, Icelandic glaciers have experienced significant mass loss, and numerous paraglacial slope failures have been documented in the country. One such failure occurred in late February 2013, when a large landslide fell onto the Svínafellsjökull outlet glacier in southeast Iceland. Digital elevation models and aerial imagery were used to quantify the glacial and paraglacial changes leading up to the event, reconstructing the processes that occurred during the landslide and the effects of the debris on the glacier surface. Between 1994 and 2013, glacier thinning and glacier‐retreat exposed a steep lateral moraine perched on bedrock which later failed and caused the landslide. Increased pore‐water pressure after an intense rainstorm and potential fluvial erosion at the toe of the source area are considered to be the primary trigger mechanisms. Morphological evidence indicates multiple phases of movement in the source area and a highly water‐rich debris avalanche on Svínafellsjökull. The debris reached a runout distance of almost 4 km and affected an area of about 1.7 km2. The estimated displaced volume of the slide is 5.33 ± 0.08 × 106 m3, making it the largest documented landslide originating from unconsolidated material in Iceland. The glacier surface ablation beneath the debris deposits was reduced due to insulation, whereas increased glacier thinning was observed surrounding the deposits, resulting in an up to 35 m height difference between the debris‐free and the debris‐covered ice by 2020. This study shows that large catastrophic landslides can originate from, and result in the formation of ice‐cored or frozen sediment complexes and highlights the potential risk coming from similar slopes around the world as glaciers continue to recede and the number of paraglacial landslides increases.


| INTRODUCTION
Large gravitational mass-movements often occur within the paraglacial period of enhanced geomorphological activity conditioned by deglaciation (Ballantyne, 2002a;Porter, Smart, & Irvine-Fynn, 2019) in both rock and sediment slopes (Ballantyne, 2002b). Retreating valley glaciers often leave behind lateral moraines which are deposited on glacially eroded bedrock. These moraines can be extremely steep (up to around 70 ) and can remain stable for decades after deglaciation (Curry, Sands, & Porter, 2009) but may become unstable at a certain threshold (Blair, 1994).

Most glaciers on the south coast of Iceland reached their Little
Ice Age (LIA) maximum extent around 1890 (Björnsson, 2017;Evans, Archer, & Wilson, 1999;Gudmundsson, 1997;Gudmundsson, Björnsson, & Pálsson, 2017;Þ orarinsson, 1943;Sigurdsson, 2005). After the LIA, the glaciers retreated at a relatively constant rate until 1970-1980 when a period of slow but significant glacier advance started, lasting until 1994 Sigurdsson, 2010). Since 1994, Icelandic glaciers have been retreating at varying rates, with the highest rate of ice loss between 1994 and 2010, and mostly slower rates since then . Observations from the Stykkish olmur weather station show a temperature rise of about 0.7 C over the 20th century and an increase of about 0.5 C per decade between 2000Björnsson et al., 2018;J onsson et al., 2014). During this time, summer air temperatures appear to have been the strongest driver of glacier recession in the southeast region (Bradwell, Sigurdsson, & Everest, 2013;Chandler, Evans, & Roberts, 2016). Iceland's average annual precipitation increased from 1500 mm/yr to 1600-1700 mm/yr between 1980 and 2015 (Björnsson et al., 2018). Measurements from glacial rivers in Iceland show mechanical erosion rates of up to 10,200 t/km 2 /yr which rank among the highest documented values worldwide (Louvat, Gislason, & Allègre, 2008). These high rates are attributed to the presence of relatively young, glassy, volcanic bedrock in heavily glaciated terrain. Gravitational mass movements are ubiquitous in Iceland and pose a significant threat to settlements and other infrastructure in almost every part of the country (Bell & Glade, 2004;Ilmer et al., 2016). Over the last decades, several large paraglacial slope failures of different types have been reported in Iceland (Table 1) and their timing generally coincides with periods of glacial retreat (Hanna, J onsson, & Box, 2004;Hannesd ottir et al., 2015;Saemundsson & Margeirsson, 2016). When looking at large slope failures (> 10 5 m 3 ) in several high mountain regions of the world there seems to be a trend of increasing frequency in the last three decades compared to the previous part of the 20th century (Fischer, 2009;Huggel, Clague, & Korup, 2012;Liu, Wu, & Gao, 2021). This trend seems to be true for Iceland too, but must be further investigated to avoid statistical bias caused by more reported landslides as a result of improving remote sensing techniques.
In line with the apparent trend of increasing paraglacial slope instabilities, a landslide in late February 2013 resulted in a large debris avalanche onto Svínafellsjökull, an outlet glacier on the western slopes of the Öraefajökull volcano, southeast Iceland (Figures 1 and 2). The landslide occurred after a strong rain event and involved a large debris volume which covered a significant part of Svínafellsjökull glacier surface, but hitherto there have been no published studies of the event or the long-term effects of the debris on the glacier. We present the investigation of the causes and effects of the landslide as a processbased quantification study of paraglacial activity and its implications for geomorphic change and landslide hazard in freshly deglaciated terrain.

| GEOGRAPHIC AND GEOLOGIC SETTING
Iceland is located in the North Atlantic Ocean just below the Arctic Circle. The country's landscape is dominated by volcanoes and glaciers and is exposed to a cool North Atlantic maritime climate. Located in the southeast of the country, Vatnajökull is Iceland's largest ice cap with 28 temperate outlet glaciers distributed around its margins (Björnsson, 2017). The Öraefajökull central volcano is covered by an ice cap that forms the southernmost part of Vatnajökull. The summit plateau is just over 1800 m above sea level (a.s.l.), and north of it, Hvannadalshnjúkur peak (2110 m a.s.l.) forms Iceland's highest summit ( Figure 1).
Öraefajökull ice cap has the highest mass-balance sensitivity to changes in temperatures and precipitation in Iceland . Adding to the complexity of the location, the volcano has produced two major eruptions in 1362 and 1727 which led to tephra fall, earthquakes, and extensive ice melt resulting in jökulhlaups (Th orarinsson, 1958;Roberts & Gudmundsson, 2015).
One of the outlet glaciers on the western slope of Öraefajökull ice cap is Svínafellsjökull, which emerges from the plateau of the ice cap and runs down a 1000 m high, steep, and heavily crevassed slope between Hvannadalshnjúkur and the peaks of Mount Hrútsfjall. The glacier is aligned northeast to southwest and flows for about 7 km through a steep valley confined by Mount Svínafell to the south and Mount Hrútsfjall to the north. The ablation area of Svínafellsjökull has lowered between 60 and 100 m during 1945-2011  and formed two proglacial lakes at the glacier snout (Gudmundsson et al., 2020).
Below the northeast face of Skarkðatindur peak (the highest summit of Mount Svínafell) the Dyrhamarsjökull outlet glacier, a former tributary glacier, flows towards Svínafellsjökull from the southeast.
The source area of the slide is located at the former confluence of Svínafellsjökull and Dyrhamarsjökull at the area around N64.0131 , E16.785 (red star in Figures 1 and 2). Due to its location just below the steep northeast face of Skarðatindur peak (Figure 3) (Helgason & Duncan, 2013). East of Skarðatindur peak, an intrusion of rhyolitic/dacitic composition has been described in a geological map by Roberts and Gudmundsson (2015). In their map, the source area of the landslide addressed in this study is marked as 'Holocene sediments and tephra on flanks'. This formation is deposited on top of the tertiary basaltic lava layers identified by Helgason and Duncan (2013).
Since the LIA the valley which hosts Svínafellsjökull has been undergoing rapid changes Everest et al., 2017;Gudmundsson et al., 2020). Not only has Svínafellsjökull retreated and thinned substantially, but gravitational mass movements such as rock fall and serac collapses are daily occurrences in the valley. Geomorphological analysis of proglacial moraines of three neighbouring outlet glaciers indicates that Svínafellsjökull has had comparatively high supraglacial and englacial sediment input (Thompson, 1988;Lee, Maclachlan, & Eyles, 2018), which suggests higher gravitational mass-movement activity in the valley. This collection of evidence highlights the fast rate at which the paraglacial environment at the outlet glaciers of Öraefajökull ice cap, and particularly in the Svínafellsjökull valley, adjusts to glacier retreat.
The glacial outwash plains around Öraefajökull volcano experience a relatively mild oceanic climate with a mean annual temperature range of about 11 C and around 150 days of precipitation per year (Einarsson, 1984;IMO, 2020). Average temperatures during winter months are typically around 0-4 C, and daily minima are rarely lower than À5 C at Fagurh olsmýri weather station, approximately 20 km southeast of Svínafellsjökull. Mean annual precipitation is around 1800 mm south and west of Öraefajökull volcano (Bradwell, Sigurdsson, & Everest, 2013). However, the precipitation increases dramatically on the slopes of Öraefajökull volcano with an annual average of up to 10,000 mm (water equivalent) at the summit plateau (Crochet et al., 2007). The source area of the slide described in this article is located halfway between the glacial outwash plain and the summit of Öraefajökull volcano, and therefore the annual precipitation at the site is expected to be between 1800 and 10,000 mm/yr.

| METHODOLOGY
Most of the investigation was carried out using a range of remote sensing data sources (Belart et al., 2019;J ohannesson et al., 2013;Porter et al., 2018;Sigmundsson, 2019). The details of the digital elevation models (DEMs) used in this study are listed in Table 2.
All volume related calculations were performed in QGIS 3.16.1 and Python 3. Before the volume was calculated, the 2013 DEM was georeferenced to the 2011 DEM in order to align the two raster files.
The 'Georeferencer Tool' was used to perform a linear shift based on five points around the source area, which were clearly identifiable in both raster files.
Two differential digital elevation models (dDEMs) were calculated by subtracting two DEMs of different years from one another to visualize elevation changes over time. One dDEM was produced by subtracting the 2011 DEM from the 2020 DEM to visualize the longterm effect of the debris cover on Svínafellsjökull and is hereafter referred to as 'dDEM11-20'.
The calculation of landslide volume and related parameters was performed by generating a dDEM between 2011 and 2013, hereafter referred to as 'dDEM11-13'.
The source area ( Figure 4) was defined by a polygon outlining the area where material was removed based on dDEM11-13 ( Figure 5). Pléiades imagery (Sigmundsson, 2019) The volume (V) was calculated (Equation 1) by multiplying the average negative elevation change dh À Á in the source area by the size of the source area (A s ): The uncertainty was calculated (Equation 2) by using the normalized median absolute deviation (NMAD) approach explained in Höhle and Höhle (2009) and Berthier et al. (2014).
where Δh j stands for the individual errors and m Δh denotes the median of the errors.
The elevation data from 2011 was subtracted from the 2013 data in an area unaffected by the slide right next to the source area with a similar aspect direction and slope angle ( Figure 4). After the DEM co-registration, the average elevation difference in stable areas should be zero. With the value for NMAD, we can calculate (Equation 3) the volume uncertainty ΔV: To compare the runout of the landslide with other events, the Heim's ratio μ H (a.k.a., apparent friction) was calculated (Equation 4) (Lucas, Mangeney, & Ampuero, 2014;Parez & Aharonov, 2015;Peruzzetto et al., 2020) using the ratio: where H is the total drop height, L is the total horizontal travel distance along the flow path and a is the angle of reach. These numbers were measured from the respective elevation data in QGIS 3.16.1 along the profile E-F in Figure  Data from a seismic station about 5 km southwest of the source area was inspected but did not show any sign of landslide-triggered seismic signal during the timeframe in question and is therefore not included in this study.

| Dimensions of displaced material
Based on the calculations described in Section 3, the respective parameters of the source area were computed and are shown in Table 3.
The calculated volume removed from the source area is

| Morphological analysis
In order to understand the processes during the slide, the pre-and post-landslide morphology of the source area (Figure 6a

| INTERPRETATION AND DISCUSSION
In this section we reconstruct the 2013 landslide on Svínafellsjökull and discuss its precursors, triggering factors, the effects on the glacier surface, as well as its implications for paraglacial slopes in Iceland and similar settings around the world.

| Origin of the source material
The presented data indicate that the failed sediment complex consisted of lateral moraine and overlying colluvial material from Skarðatindur peak which were deposited upon a northeast dipping, glacially eroded, bedrock slope.
The lateral moraine was created by Dyrhamarsjökull when it was still connected to Svínafellsjökull. The age of the lateral moraine is not clear, but due to its position and size when compared to other south coast glacier forelands, a LIA maximum age is likely (Bradwell, Sigurdsson, & Everest, 2013;Evans, Ewertowski, & Orton, 2017a, 2017b, 2019Chandler et al., 2020).

| Long-term destabilization of the slope
When comparing the DEMs in the profile in Figure 6(a), it becomes evident that Dyrhamarsjökull retreated and thinned for decades before the slide occurred. Between 1994 and 2011, the moraine complex which forms the source area of the discussed event became exposed as the ice of Dyrhamarsjökull thinned by up to 4.7 m/yr, creating a steep (46 ) and unstable slope. This was a period of dramatic ice loss on all south Icelandic glaciers (Adalgeirsd ottir et al., 2020), creating unprecedented rates of moraine abandonment and paraglacial adjustment, especially within the mountainside reaches of glacier snouts (Bennett et al., 2010;Bennett & Evans, 2012). This rapid exposure of the moraine complex at Dyrhamarsjökull suggests a lack of response time for the moraine to incrementally adjust to the deglaciation.
Since a large number of blocks of highly debris-rich glacier ice were observed in the landslide debris (Figure 10b), it is likely that the lateral moraine was at least partly ice-cored, or permafrost preserved from the LIA was present within the sediment. Thawing of the ice within the moraine due to its recent rapid exposure would have exacerbated its destabilization (Holm, Bovis, & Jakob, 2004;Tonkin et al., 2016;. Additionally, continuous rockfall from the steep northeast face of Skarðatindur peak has led to the accumulation of the colluvial material on top of the moraine, thereby increasing the normal load on the slope. The location of the source area at the former glacier confluence means that the sediment complex became exposed from two sides when Dyrhamarsjökull and Svínafellsjökull receded, making it more vulnerable to erosive forces than a straight lateral moraine. Therefore, we assume that there was an ongoing rain event in the landslide source area that reached its precipitation peak during the storm 24-48 h before the slide occurred. Since the slide occurred in late February, the rain in combination with positive temperatures at up to 2000 m a.s.l., led to significant snow melt, which further contributed to the total runoff. This almost certainly infiltrated the sediment in the source area, increasing the pore pressure and consequently destabilizing the slope (Iverson, 2000). A concomitant increase in volumetric water content would also have added weight to the sediments, further favouring failure (Iverson, 2000). Percolation through the sediment to the contact with underlying basaltic bedrock would also have resulted in increased basal water pressure, leading to reduced shear strength at the soil-rock interface and thereby creating a sliding plane. The calculated catchment and drainage channels ( Figure 9) suggest that surface water from an area of 3.2 km 2 may have drained through a lateral meltwater channel immediately below the toe of the sediment complex (bold dashed line in Figure 9). Since the calculations only take surface runoff over Dyrhamarsjökull into account and englacial or subglacial hydrology is disregarded, the model presented here is a simplification but can be used as an approximation to understand where meltwater channels would have formed.

| Short-term destabilization of the slope
If this channel formed at the location where our model calculations predict it, a significant amount of water (rainwater + meltwater) drained just below the sediment complex and potentially undercutting the slope, destabilizing it further.

| Timing
Witness statements from Skaftafell indicate that the slide occurred around 20:00 h on 27 February 2013. However, there is no proof for the exact timing, and the possibility remains that the event happened earlier during the rainstorm. Assuming the reported timing of the slide is correct, the slide occurred more than 24 h after the peak of the rainstorm. This reflects the infiltration duration and subsequent elayed peak pore-water pressure within the sediment as described by Iverson (2000).  Figure 6b).

| Progression of the slide
After the lower part of the source area became mobilized, the rest of the sediment complex likely failed retrogressively, forming the terraces. The morphology of the terraces suggests two phases of slope deformation in different directions. The yellow arrows in Figure 10(a) indicate the first phase where the material slid down the bedrock slope below Skarðatindur peak towards the northeast, largely maintaining its internal structure.
During the second phase, the terraces were formed by rotational slumping of sediment blocks (Varnes, 1958;Hungr, Leroueil, & Picarelli, 2014) towards the northwest along at least three failure surfaces. This motion is illustrated by the black arrows in Figures 10(a) and 12. Figure 12 shows a cross-section of the geometry of the sliding surface and the original pre-slide surface as a red line. Prior to the landslide, the slope in the same area dipped towards the northeast, and the post-slide terrace surface is flat, indicating that most of the removed volume was in the western part of the source area. The fact that the terrace surface is a preserved pre-slide surface means that a significant amount of material (30-40 m vertical) must have been removed from underneath the terraces ( Figure 12).
Quantitative data collected about the landslide (Heim's ratio, displaced volume, horizontal travel distance, vertical fall height and affected area) compared to different landslide databases (Legros, 2002;Lucas, Mangeney, & Ampuero, 2014;Parez & Aharonov, 2015) indicates that the event classifies best as a debris avalanche. This is further supported by the wide distribution of the debris on Svínafellsjökull and channelized systems within the deposits (Zhou et al., 2019). Given that the source material at the time of slide initiation had a high water content, it is likely that debris avalanche formed shortly after the initiation of the slide. During its progression, the debris likely entrained dead ice from the source area, glacier ice, surface water, and snow along the flow path, further increasing the water content (Evans & Delaney, 2015).
The fact that the debris was flowing over glacier ice is likely to have reduced frictional resistance and further increased the runout distance and deposition area (Iverson, 1997;Evans & Delaney, 2015).

| Effects on the glacier surface and implications for dead-ice environments
Between 2013 and 2020, the landslide debris was transported around 1000 m with the glacier flow ( Figure 4). In that time period, a distinct F I G U R E 1 2 Simplified schematic model of proposed sliding plane from profile C-D in Figure 6(b). Arrows indicate sliding direction. The 2011 and 2013 surfaces are taken from the elevation models [Color figure can be viewed at wileyonlinelibrary.com] elevation difference of up to 35 m developed between the bare glacier surface and the debris-covered glacier, similar to that described from rock avalanches deposits on glaciers in Iceland (Saemundsson et al., 2011) and in Alaska (Shreve, 1966;Shugar et al., 2012). This shows clearly that the debris is insulating the underlying ice very effectively as reported by Reznichenko et al. (2010) and Saemundsson et al. (2011). The observed increased lowering of the immediately surrounding glacier ice (Figure 11) can be explained by the effect of increased surface melt due to a thin dust layer (Oerlemans, Giesen, & Van Den Broeke, 2009;Deline et al., 2014;Fyffe et al., 2020) derived from the landslide debris. Energy from solar radiation stored in the debris cover is being reemitted as long wave radiation (Cuffey & Paterson, 2010) and might lead to additional increased melting in the surrounding. The extent to which this sensible heat flux from the debris body on the glacier ice contributes to the augmented ablation rate in its vicinity must be tested in further studies.
These processes are likely to continue as the landslide material is transported towards the frontal glacier margin where they are going to accelerate the isolation of the debris covered glacier leading to incremental stagnation and the formation of a dead-ice environment.
Therefore, the results of this study show that ice-cored/frozen sediments can be the origin and result of catastrophic paraglacial landslides. The evolution of dead-ice environments has been studied thoroughly at Icelandic outlet glaciers (Kjaer & Krüger, 2001;Bennett et al., 2010;Bennett & Evans, 2012) but future studies on the evolution of landslide debris covered glaciers and the surroundings at Svínafellsjökull and Morsárjökull would help to understand occurring processes better. According to Thompson (1988) and Lee, Maclachlan, and Eyles (2018) the frontal moraine of Svínafellsjökull is characterized by a high percentage of supraglacial and englacial material which suggests that increased landslide activity at Svínafellsjökull has been ongoing throughout the formation of the moraine. The debris from the 2013 landslide will eventually be added to this moraine supporting this assumption.

| Implications for paraglacial slopes
Iceland's glaciers have experienced significant ice loss since the LIA, a trend that will continue for the foreseeable future (Adalgeirsd ottir et al., 2020). An increased paraglacial response during this deglaciation period is therefore to be expected (Ballantyne, 2002a). In light of numerous gravitational mass movements and slope instabilities in Iceland during the last decades (Kjartansson, 1967;Sigurdsson & Williams, 1991;Saemundsson et al., 2011), the 2013 landslide onto Svínafellsjökull confirms an apparent trend of increased paraglacial landslide activity in Iceland.
The event addressed in this study shows an extreme example of the modification of a recently deglaciated sediment-mantled slope.
The process of moraine wasting through gravitational mass movement is well known in the literature (Ballantyne, 2002b;Ravanel et al., 2018). Large mass movements of lateral moraines have been documented worldwide (Blair, 1994;Hewitt, 2013;Kumar et al., 2019;Cody et al., 2020) and are generally associated with rapid glacier retreat and thinning. Most failures involving lateral moraines discussed in the literature are slow-moving, which either indicates that catastrophic failures of lateral moraines are rare or that they are seldom documented. The large volume and runout involved in the catastrophic landslide in 2013 is rather unusual and likely the result of the combination of factors discussed in this article. Rapid lowering of the glacier surface and a potentially ice-cored moraine likely did not allow for gradual adjustment of the lateral moraine, which led to the catastrophic failure in combination with intensive rainfall, snowmelt, and potential fluvial undercutting of the slope. The proximity of the landslide source area to the snout of Dyrhamarsjökull indicates that the observed rapid glacier surface lowering was due to a combination of glacier retreat and surface ablation. Other paraglacial slopes at similar relative locations to glacier margins (e.g., Cody et al., 2020) may undergo a comparable loss of lateral support, and are thus likely to experience some form of increased paraglacial response. Blair (1994) observed a critical relief height between glacier surface and moraine crest at which sediment mantled slopes above Tasman Glacier, New Zealand, developed complex slope failure. The possibility of such a site-specific critical relief height in sediment-mantled slopes which are exposed by glacial thinning could indicate that large amounts of sediment stored on rock slopes will become unstable at a certain point during or after deglaciation. In 2020, lateral moraine crests were between 50 and 90 m higher than the Svínafellsjökull glacier surface.
With continuing deglaciation of the valley these moraines will become further exposed and failure will become more likely. In the last decade, observed a three-to four-decade long trend of permafrost degradation in certain regions in Iceland and predicts that this process will continue as temperatures increase further. Increasing exposure of paraglacial slopes, in combination with more frequent and more intense rain events and regional permafrost thawing, will undoubtably increase the occurrence of paraglacial landslides in Iceland in the future. Therefore, it is necessary to monitor glacier margins regularly to track the exposure of potentially dangerous rock and sediment slopes.

| CONCLUSIONS
The landslide which fell onto the Svínafellsjökull in 2013 is an extreme example of paraglacial sediment-mantled slope adjustment. It is the largest documented landslide originating from unconsolidated paraglacial material in Iceland with a displaced volume of 5.3 AE 0.08 Â 10 6 m 3 and a runout distance of 3.95 km. The source material originated from an at least partly frozen or ice cored sediment mantled slope.
The sediment mostly consisted of lateral moraine material with accumulated colluvium on top, located just below the northeast face of Skarðatindur peak. Our data clearly shows that the slope had been exposed over several decades prior to 2013 by retreat and thinning of Dyrhamarsjökull, which made it vulnerable to gravitational erosion and highlights the especially high landslide hazard just at the retreating frontal margin of glaciers.
Increased pore-water pressure within the sediment and at the bedrock interface caused by rain and snow melt after an intense rainstorm event is thought to be the main triggering factor of the slide.
Our analysis indicates that the landslide was likely initiated at the toe in the northern part of the source area as a retrogressive debris slide, which led to a subsequent phase of sliding motion and multiple rotational failures that formed prominent sediment terraces in the southern part. During the runout over Dyrhamarsjökull and Svínafellsjökull, the debris slide transformed into a debris avalanche. The debriscovered part of the glacier is efficiently insulated, and the surrounding debris-free glacier shows increased ablation rates which resulted in an up to 35 m high elevation difference from the debris-free surface by 2020. These processes will likely accelerate the evolution of a deadice environment once the debris has been transported to the frontal glacier margin. The described event is an indicator for increased recent paraglacial activity at Iceland's outlet glaciers and especially in the Svínafellsjökull valley where a cluster of slope instabilities is observed. This trend is likely to continue as deglaciation proceeds and heavy rain events are predicted to increase.
Finally, this landslide highlights the range of paraglacial slopes which can be affected by catastrophic failure and emphasizes the need to further investigate and monitor paraglacial sediment structures.