A number of suspected glacier detachments and ice–rock avalanches happened around Mount Kazbek at the border between Russia and Georgia. During the 18–19th centuries, surges of the Devdorak (Georgian name Devdoraki also used in literature) Glacier on the north-eastern flank of Mount Kazbek (Fig. 2; Table 1) were moving down the Amilishka River valley (Kabakhi River in its lower part), which drains the glacier catchment. Surge-like advances were recorded in 1776, 1778, 1785, 1808, 1817, and 1832 (Zaporozhchenko and Chernomorets, 2004). On 13 August 1832, parts of the surging Devdorak Glacier tongue detached and the subsequent ice–rock avalanche blocked the main Terek valley, an important transportation route between Russia and Georgia (Petrakov et al., 2008). An eyewitness of the 1832 event, Engineer-colonel Grauert, estimated a volume of 15.5 × 106 m3 of ice and rock mass blocked the Dariali Gorge of the Terek valley (Zaporozhchenko and Chernomorets, 2004). The causes and mechanisms of this (and other) detachments are not well known and vary throughout the literature. Ice–rock avalanching onto the glacier, overloading, and the associated increase in subglacial water pressure could well have played a role.

Except for their source area, these at their time well-known “Kazbek blockages” followed the same avalanche path as the 2014 Devdorak event described in the following. In 2014, parts of a rock wall and overlying hanging glaciers failed from Mount Kazbek (Figs. 2–3). The resulting highly mobile ice–rock avalanche of 2–5 × 106 m3 rushed down the Amilishka and Kabakhi valleys and blocked the main road between Russia and Georgia, killing nine people (Chernomorets et al., 2016; Tielidize et al., 2019). This event might also have had a longer runout (Fig. 1) but was stopped by a sharp 90∘ turn at the confluence of the Devdorak gorge and the Dariali (or Terek) gorge. The 1832 Kazbek blockage is described as a glacier detachment in the sense of the present contribution in older literature, but more recent interpretations, not least based on the 2014 event, indicate that the 1832 event and other such events at the site might have started as rock failures rather than surges (Chernomorets, 2014; Chernomorets et al., 2016). We list the Kazbek blockages as potential glacier detachments here but stress that it remains uncertain what kind of event they actually were.

Currently, the entire narrow lower part of Devdorak Glacier has a surface slope of around 23∘, its tongue closer to 17∘. The volcanic nature of Mount Kazbek, the documentation of a number of violent mass flows from the mountain in the past (Chernomorets et al., 2007), field visits, and visual analysis of very high-resolution satellite images and terrestrial photos all indicate highly erodible rock and an abundance of fine sediments at many places on Mount Kazbek. In all high-resolution satellite images available since around 2002 (GoogleEarth, Bing Maps, Maxar, Pleiades), Devdorak Glacier appears heavily crevassed and is partly covered by fine sediments, likely deposited by mass movements from the surrounding mountain flanks. Since 2015, a glacier directly south of Devdorak Glacier has been overriding the main tongue of Devdorak glacier in a surge-like destabilization, perhaps triggered by the 2014 Kazbek/Devdorak ice–rock avalanche that overran it (Fig. 3). Possibly as a consequence of the tributary surge, Devdorak Glacier itself is currently also advancing (Dokukin et al., 2020). A 1 km global permafrost model (Obu et al., 2019) not particularly tuned for mountain permafrost indicates the elevation of the Devdorak Glacier tongue is roughly at or below today's lower boundary of the discontinuous permafrost zone in the region.

The 130 × 106 m3 Kolka Glacier detachment (Figs. 2 and 4; Table 1) of 20 September 2002 has been described and discussed in several studies (Kääb et al., 2003; Haeberli et al., 2004; Kotlyakov et al., 2004; Huggel et al., 2005; Drobyshev, 2006; Evans et al., 2009b). During these investigations, it became clear that a similar event must have already happened at least once, at the beginning of July 1902 (and probably also around 1700), whereby the glacier tongue detached after a roughly 2-week-long surge-like advance. Heavy rain and snowmelt may have played a role in triggering the 1902 advance. Damming or erosion by water could then have caused the actual detachment and subsequent rock, mud, and/or ice flow (Drobyshev, 2006; Petrakov et al., 2008; Kotlyakov et al., 2010b). The descriptions given in these references and their sources clearly describe an event that qualifies as a detachment in the sense of the present contribution. The glacier started surging again in autumn 1968 and advanced by 4 km with speeds of up to 220 m d-1 (Fig. 4) but without catastrophic consequences (Hoinkes, 1972; Rototayev et al., 1983). We do not know which differences in conditions, compared to 1902 and 2002, caused the glacier to not detach in 1970. Several other glaciers in the Caucasus are known to have surged in the past (Kotlyakov et al., 2010b). We found that Kolka Glacier had a low surface slope of around 13∘ prior to the 2002 detachment. Evans et al. (2009b) estimated a bed slope of 9∘ after the detachment. Over the course of several weeks before the 2002 detachment, perhaps triggered by earthquakes (Kotlyakov et al., 2004), heavy rock and ice falls from the northern flank of the Kazbek massif deposited several 106 m3 of material on the glacier. During this period of mass-wasting activity, the glacier changed in unusual ways: it bulged and became heavily crevassed (Fig. 4), it developed a scarp at the location of the later detachment, and supraglacial ponds formed (Kotlyakov et al., 2004, 2010b; Evans et al., 2009b). Unusually high geothermal heat fluxes underneath the glacier (fumaroles and a sulfur smell were reported from the glacier bed shortly after detachment), the impact energy of a large rock and/or ice fall and successive loss of shear stress due to excess water pressure have been proposed as possible factors and ultimate triggers of the 2002 detachment (Kotlyakov et al., 2004, 2010b; Evans et al., 2009b). Similarly, the additional loading of Kolka Glacier from the rock and ice falls has more recently been proposed to have increased the basal shear stress until it exceeded a frictional threshold given by the glacier bed material, topography, and hydraulic conditions (Kääb et al., 2018). After the detachment, lakes were visible on the Kolka Glacier bed, pointing to the involvement of large amounts of subglacial water in the detachment. The high mean avalanche velocities of 50–80 m s-1 (Huggel et al., 2005) suggest the availability of large amounts of water and/or saturated fine-grained till.

The Kolka Glacier tongue is, like Devdorak Glacier, roughly at the lower elevation of the regional discontinuous permafrost zone, and the glacier must have been temperate throughout, except for the likely polythermal or even cold steep hanging glaciers avalanching from the northern flank of the Kazbek massif onto its surface. The ice–rock avalanche resulting from the 2002 detachment is described in detail in the above literature about the event, but we want to draw attention to the streamlined debris stripes, in the following called debris stripes, that were visible in the detachment zone after the event (Fig. 4d; Petrakov et al., 2004; Huggel et al., 2005) because similar patterns have also been found in several of the other cases of this study. Currently, the detached glacier depression is refilling and the glacier is gaining mass again. The fast recovery of the glacier volume lost by the 1970 surge, and again after the 2002 detachment, is associated with Kolka Glacier's positive mass balance, which stands in stark contrast to the predominant strong glacier shrinkage in the Caucasus Mountains (Kutuzov et al., 2019; Zemp et al., 2019). Kolka Glacier had already reached almost 50 % of its pre-detachment volume by 2017 (Supplement Fig. S2) and is projected to accumulate 60 %–70 % of its pre-detachment volume by 2025 (Petrakov et al., 2018; Aristov et al., 2019).

The 2017 event, first mentioned in Dokukin et al. (2019), happened between 10 and 11 July 2017 (Planet images). A glacier of roughly 1000 m length and 240 m width detached (upper scarp ca. at 3600 m a.s.l.), and the resulting ice–rock avalanche flowed down a narrow valley towards the village of Tojikobod in the Rasht Valley (Figs. 5 and 6; Table 1). In satellite imagery taken shortly after the event (Planet, Maxar), ice remains can be recognized over a horizontal distance of about 8 km, down to an elevation of about 2160 m a.s.l. Beyond this point, a considerable debris and/or mud flow must have continued for another 2 km or so. The detached glacier had a surface slope of around 16∘ (High Mountain Asia digital elevation model, DEM; Shean, 2017). We measured increased surface speeds of up to 5.8 m d-1 in early July 2017 (compared to a few decimetres per day, dm d-1, in 2016; velocities from repeat Planet data) and detected unusual lateral crevasses delineating the later detachment area in images from as early as May 2017 (Fig. 6). The glacier is likely not surrounded by permafrost (Obu et al., 2019).

To roughly estimate the event volume, we derive the ice thickness along the centre flow line of the glacier based on an estimated basal shear assuming a driving stress of 1.2 × 105 Pa as suggested for mountain glaciers, a slope of 16∘, and a form factor of 0.8, which then results in a thickness of around 60 m (Cuffey and Patterson, 2010). Multiplying half of this depth (i.e. assuming a triangular cross section) with the detached area (ca. 200 000 m2) gives a first volume estimate of roughly 6×106 m3. Whereas satellite images after detachment suggest that much of the glacier bed might actually have a triangular cross section, it may have been more shallow in the lowermost and uppermost parts. As an order of magnitude, we suggest a detachment volume of 5 × 106 m3 and assign a conservative error of ±1 × 106 m3 to this estimate.

The cirque from which the 2017 event originated was also the source of other slope instabilities over recent years. Another (much smaller) ice–rock avalanche from a neighbouring glacier occurred between 15 and 24 July 2016 (dates from Planet images), with a horizontal reach of roughly 5.5 km. A large debris flow descended the same valley in late August 2016, starting from the same cirque, likely entraining deposits of the July 2016 avalanche and reaching the Surkhob River 19.4 km downstream where it destroyed several buildings, bridges, and agricultural fields.

Between 2 and 3 August 2019 a second glacier, ca. 14 km to the west of the one that collapsed in 2017, detached (Figs. 5, 7). This glacier was slightly smaller than the 2017 one and had a surface slope of around 20∘. The resulting ice–rock avalanche travelled north down a narrow valley towards the Rasht Valley over a horizontal reach of 6.5 km and a vertical drop from 3350 (upper scarp) to 1850 m a.s.l. Superelevations of the ice–rock avalanche path of up to almost 200 m above the valley bottom suggest high avalanche speeds (McClung, 2001). We estimated the detached volume in the same way as described for the 2017 event (Sect. 3.2.1) and computed a volume of 4.5±1 × 106 m3 (glacier area ca. 230 000 m2; centreline depth 50 m). Based on pre- and post-event WorldView stereo DEMs, Leinss et al. (2020) estimate a maximum erosion depth of 90 m and a detachment volume of 8–9 × 106 m3, i.e. significantly more than our rough general model.

This glacier also showed increased sliding speeds and crevassing around the later detachment area for at least 2–3 weeks before the failure. For the end of July 2019 we found surface speeds of roughly 2.5 m d-1, a marked increase compared to roughly < 0.1 m d-1 during 2017–2018 (repeat Planet data). The glacier did not show any visual signs of destabilization between 2015 and 2018 (Planet images). In 2007 (Maxar; see Supplement Fig. S3) the glacier looked heavily crevassed, possibly an indication of a surge-like advance. This condition is still visible in Landsat data 5–6 years later, though less certain due to the lower resolution of Landsat data (no other data are available to us between 2007 and 2015). Landsat data also suggest that the glacier experienced a similar advance in the early 1990s. Under the limitation of the reduced spatial resolution of the Landsat data, however, we do not find signs of a large detachment event or large ice–rock avalanche. Nevertheless, we draw attention to a surprisingly vegetation-free landform visible downstream of the 2019 detachment in pre-event imagery (Fig. 8). The lack of vegetation, the streamlined microtopography, and zones of rough and chaotic microtopography that resemble avalanche or debris flow deposits (Fig. 8; Supplement Fig. S4) led us already before the 2019 event to interpret this landform as a possible geomorphological imprint of an earlier ice–rock avalanche (Kääb, 2019). Meanwhile, this landform has been overrun by the August 2019 ice–rock avalanche, leaving similar new forms, and suggesting that the landform now buried could have originated from a similar detachment event, probably before 1961 (year of earliest Corona satellite image). Between the 1961 Corona images and very high-resolution images from just before the 2019 avalanche, no significant changes are visible on the landform.

Very high-resolution satellite images over the Peter the First Range suggest an abundance of weak bedrock and fine sediments. All over the range, signs of large debris flows, rock avalanches, or ice–rock avalanches are visible in high-resolution satellite images (Maxar, CNES/Airbus, Planet; Leinss et al., 2020). The Pamirs are known to be very geomorphologically active, with a number of associated hazards (Mergili et al., 2012; Gruber and Mergili, 2013; Strom and Abdrakhmatov, 2018) and a cluster of surge-type glaciers (Kotlyakov et al., 2008, 2010a; Gardelle et al., 2013; Sevestre and Benn, 2015; Lv et al., 2019; Goerlich et al., 2020). A number of glaciers in the Peter the Great Range were surging at the time of writing or have done so in the recent past (Fig. 5). Didal Glacier (ca. 12∘ steep) surged around 1995 and again during the winter 2015/16 when it advanced by 2.5 km over a few months. In the valley below Didal Glacier we note a lack of vegetation in the valley and landforms that could well stem from a former ice–rock avalanche. Kotlyakov et al. (2010a) mention a 2.2 km long ice avalanche from Didal Glacier in 1974. In comparison, the bottom of the valley through which the 2017 ice–rock avalanche descended (Sect. 3.2.1) was partly tree covered before, suggesting that no event like the 2017 one has happened there in the recent past.

Even if not documented in detail in an internationally accessible format so far, to our best knowledge, both the 2017 and 2019 detachments and their downstream effects were very likely noted by the local communities as the lowermost ice and rock deposits stopped not far from settlements, agricultural fields, and pastures, and very high-resolution images (Maxar) show that flooding happened close to houses, and two irrigation channel bridges were partially destroyed.

In order to elucidate significant elements of the recent mass-flow history from the basin that might have conditioned the 2018 detachment and that are not documented in detail elsewhere, we start our description with the oldest satellite data available to us. In Corona satellite reconnaissance data of 8 December 1969, a narrowing in the Yarlung Tsangpo river where the Sendongpu valley joins the river points to deposits from previous mass flows. However, trees on these deposits suggest no recent large mass flow activity. The main glacier in the basin, here called Sedongpu Glacier (Fig. 12), shows some signs of enhanced flow such as large crevasses. In Corona satellite data of 6 November 1974, the glacier has advanced some 800 m into steeper terrain. Fresh traces of a large mass flow are visible between the glacier front and the main Yarlung Tsangpo river, and fresh deposits seem to have covered or destroyed the forest on the older deposits in the river, but the main glacier is still in place. The glacier showed one single tongue in 1969 but had split into two tongues during its advance by 1974, a feature that it still exhibited in 2016. Chen et al. (2020) describe an ice avalanche from the Sedongpu basin that dammed the Yarlung Tsangpo river in 1968 but do not mention any other events before 2014. From our interpretation of the Corona satellite data there was either one more event between 1969 and 1974, or the event dated “1968” actually happened a few years later. The next large mass flow (listed as ice avalanche in Chen et al., 2020, and debris flow in Tong et al., 2018; we interpret at least a large debris content) happened in 2014. The source of the mass flow must have been in the upper part of the main Sedongpu Glacier or its headwall as both lateral moraines of the glacier were heavily eroded from the glacier side (RapidEye satellite data of 2013–2015). The glacier surface was not visibly changed along the glacier centre line, suggesting that the mass flow must have flowed along both glacier margins. Although difficult to determine in the satellite data available to us, the flow may have eroded the lowermost part of the glacier tongue.

Between 20 and 27 October 2017 (Planet and Sentinel-2 images) a huge rock avalanche started high up from the north ridge of Gyala Peri and ran over large parts of the Sedongpu basin and down to Yarlung Tsangpo, damming the river (Figs. 12 and 13). The event seems to have had a severe impact on Sedongpu Glacier. We generated two elevation models from 13 November 2015 Spot6 and 30 December 2018 Pleiades tri-stereo data that produced robust results despite the extreme topographic conditions. Differencing the two DEMs indicates that the October 2017 rock avalanche removed around 17 and 33 × 106 m3 of material from two close-by but separated areas, respectively (Fig. 12c). If both failures happened as part of the same event, the total volume of 50 × 106 m3 makes this one of the larger rock avalanches detected in recent decades. Based on a visual inspection of satellite data, we consider it very likely that the avalanche also involved small glaciers from the west wall of Gyala Peri and incorporated ice from the surface of the glaciers lower down as it ran over them. The Chinese seismic database registered two large “landslide” events on 22 October 2017: a M3.2 event at 06:20 (China standard time) about 16 km west of Sedongpu and a M4.0 event at 06:22 directly at Sedongpu. Chen et al. (2020) and Tong et al. (2018) confirm that at least the latter signal stems from the Gyala Peri ice–rock avalanche. Subsequent satellite images suggest that the avalanche must have changed the surface of Sedongpu Glacier drastically. It covered the glacier and much of the basin with debris and dust. A small, surge-like lobe with ponds on it appeared at the transition between the headwall and the tongue of Sedongpu Glacier (Fig. 13; GoogleEarth, Bing Maps). The eastern tributary glacier to Sedongpu Glacier also showed a surge-like lobe (Sentinel-2, Planet, GoogleEarth). Driven by the geomorphological changes in the basin, a series of debris flows, some involving ice and likely nourished from the large amounts of unconsolidated debris left behind by the ice–rock avalanche, occurred after 22 October 2017 and into 2018 (Tong et al., 2018; Chen et al., 2020). The two 18 November 2017 M5.2 and M6.9 Linzhi, Milin, and Nyingchi earthquakes, with epicentres only a few kilometres from Sedongpu, may have contributed to the triggering of the debris flows (Hu et al., 2019; Zhao et al., 2019). Lastly, we report hundreds of small earthquakes recorded under the Gyala Peri massif in 2017 and 2018, most up to M2, some up to M3, which we have not analysed further in the present study (China Earthquake Data Center, 2021, http://data.earthquake.cn/index.html, last access: 7 April 2021).

Following the 2017 ice–rock avalanche the main Sedongpu Glacier underwent drastic changes (Fig. 13). Ponds developed on its surface and along the margins (Supplement Fig. S7). Surface velocities increased from a background velocity of ∼ 0.3 m d-1 (ca. 100 m a-1) in 2017 to 1–3 m d-1 at the end of January 2018, 10 m d-1 in mid-September 2018, and 25 m d-1 in mid-October 2018 (velocities derived from offset tracking in repeat Planet, Sentinel-1, and Sentinel-2 data). The glacier surface showed several crevassed bulges (Fig. 13b) in January 2018, and progressively more crevasses appeared as the glacier tongue expanded (Fig. 13c; Supplement Fig. S7). The lower, flat glacier part separated from the steep headwall. Between 19 September 2018 (last optical image due to later cloud cover) and 13 October 2018 (Sentinel-1) the glacier advanced by almost 1 km. On 17 or 18 October (Tong et al., 2018; Chen et al., 2020), or 22:48 (China standard time) on 16 October according to Wang et al. (2020), the entire tongue of Sedongpu Glacier detached over a length of about 3.5 km (glacier width between 250 and 550 m) (Figs. 12 and 13). Wang et al. (2020) provide inversions of regional seismic signals caused by the event. Planet images of 27 October 2018 and Chinese media images confirm that large amounts of ice blocked the Yarlung Tsangpo river (Fig. 12d). Parts of the emptied glacier bed filled up with enough ice debris that another mass flow originated from there on 29 October 2018. The dam in the Yarlung Tsangpo river was estimated to be roughly 40–60 × 106 m3 in volume (Chen et al., 2020). Differencing our 2015 SPOT6 and 30 December 2018 Pleiades tri-stereo DEMs (Fig. 13d) shows two areas of distinct volume loss: around 80 × 106 m3 are missing from the main branch of the glacier and around 50 × 106 m3 from its terminus and frontal moraine. An ASTER satellite stereo DEM of 11 November 2017 suggests that the volume loss over the glacier tongue and frontal moraine cannot have occurred before November 2017. From our data we cannot tell how much of the total 130 × 106 m3 stems from the main first glacier detachment event of 17/18 October 2018 and how much from the 29 October 2018 event. However, satellite images indicate that the first event involved by far the largest volume. The detached glacier part had an overall slope of only 8–9∘. According to Obu et al. (2019) it should have been several hundred metres below the regional permafrost limit, but cold or polythermal ice is certainly found in the west wall of Gyala Peri and could theoretically be advected into the basal parts of the glacier near its tongue.

Between 19 September and 26 October 2018 (Planet) a ∼ 9 × 106 m3 rock(–ice) avalanche (volume from 2015 SPOT6 and 2018 Pleiades tri-stereo DEM differencing) originated from the south-western flank of Gyala Peri and likely reached Sedongpu Glacier (Fig. 12). The avalanche covered a small glacier below its starting zone which started a surge-like advance in early 2018. Due to insufficient satellite data (cloud cover in optical data; low resolution and radar shadow in synthetic aperture radar, SAR, data), though, we cannot tell if this avalanche happened before, during, or after the 17/18 October 2018 glacier detachment and could thus have triggered the glacier detachment. Seismic records are also inconclusive regarding the rock avalanche and the glacier detachment, with two dozen M1–2 earthquakes recorded under the Gyala Peri massif between mid September and end of October 2018.

From all the evidence collected above, it seems very likely that the 22 October 2017 Gyala Peri rock avalanche, which travelled over the Sedongpu Glacier, primed the glacier for its detachment a year later, perhaps with the additional influence of the 18 November 2017 earthquakes (Zhao et al., 2019) or one on 16 October 2018 (Wang et al., 2020), though the exact controlling mechanisms remain unclear. The effects that the ongoing mass-wasting activities from Gyala Peri had on Sedongpu Glacier could have been manyfold: additional loading on the glacier could have increased normal and shear stresses, and the 2017 rock avalanche and earthquakes could have mechanically weakened the glacier and its bed, potentially disrupting the subglacial drainage system. The large amounts of fine dust deposited on the glacier will likely have changed (enhanced?) its surface melt rates. Lastly, independent factors such as high temperatures and precipitation amounts (Tong et al., 2018; Liu et al., 2019) associated with large water input into the glacier could have complicated the changes caused by a rock-avalanche impact. The supraglacial and glacier-marginal ponds, unusual on temperate glaciers (see Haeberli et al., 2002; Kääb et al., 2004), suggest that the glacier was under high internal water pressure. The bulging ice and the strong downstream gradient in surface velocities suggest that a surge-like instability first developed in the upper section of the low-angle part of the glacier. Just how this instability propagated down-glacier, whether by exerting pressure on the lower glacier from above or by a propagation of exceptionally low basal friction values (Thogersen et al., 2019), remains unclear. High-resolution imagery and media photos of the glacier bed and detachment deposits, as well as the various debris flows that originated from the basin, suggest that the detached glacier rested on a soft bed with substantial amounts of fine material. The geology of the area is described as marble (Liu et al., 2019), which at least opens up the possibility of fine-grained sediments.

It remains to be seen to what extent the Sedongpu Glacier is able to rebuild given the strongly negative mass balances in the region (Kääb et al., 2015; Brun et al., 2017; Treichler et al., 2019; Shean et al., 2020).

The region around Sedongpu does not seem to host any obvious surge-type glaciers. However, the events at Zelunglung Glacier (29.62∘ N, 95.00∘ E; GLIMS G095018E29637N, RGI60-13.01428), 20 km south of Sedongpu, are worth mentioning. In 1950, 1968, and 1984 extraordinary instabilities propelled the glacier forward and blocked the Yarlung Tsangpo river (Zhang, 1992). The 1984 event seems to have involved only a smaller section of the glacier. Corona reconnaissance satellite images of 1969 show a massive advance (by about 4.5 km compared to 2018) but no detached glacier. However, the glacier had obviously overridden its frontal moraine, reaching almost down to Yarlung Tsangpo. Deposits, visually similar to those of ice–rock avalanches in general, cover much of the main Yarlung Tsanpgo river bed at a length of about 2.5 km downstream measured from the confluence with the Zelunglung valley (Supplement Fig. S8). Glacier advance rates of up to 1 km h-1 are reported for the 1950 event (Zhang, 1992). Such rates are far above what is typical for surges, and the glacier might have, at least in 1950, undergone an event close to a sudden detachment in the sense of the present contribution.

A detachment from a glacier on the southern flank of the Tinguiririca Volcano (Fig. 17) happened between 27 June (Landsat 5 TM) and 6 July 1994 (Landsat 5 TM). For the event, we estimate a detached glacier area of 0.2 km2. Using glacier thickness estimates derived for the 2007 case (described below), we estimate a detachment volume of roughly 4–5 × 106 m3. The climatic conditions between 1994 and 2007 were obviously favourable enough (i.e. little negative or even balanced regional glacier mass balances; Masiokas et al., 2016; Dussaillant et al., 2019) for the glacier to recover to its pre-detachment geometry.

At the same location, a glacier area of 0.46 km2 detached between 7 January (Landsat ETM+ image without collapse) and 14 January 2007 (Landsat ETM+ with collapse), producing an ice and debris avalanche of 10–14 × 106 m3 volume (Schneider et al., 2011; Iribarren Anacona et al., 2015; Figs. 17, 18). We estimate that the detached glacier had a surface slope of around 20∘. Before the 2007 detachment, the glacier's lowest elevation was at about 3500 m a.s.l., roughly at the lower regional limit of discontinuous permafrost (Brenning, 2005). The volcanic nature of Tinguiririca should be associated with weak rocks and sediments. There are fumarolic fields and hot springs within a few kilometres of the detached glacier (Pavez et al., 2016), and an inactive volcanic crater lies just a few hundred metres to the west of the detached glacier. In very high-resolution satellite images of 2007 and later (GoogleEarth, BingMaps), clear signs of hydrologic activity are visible in the upper part of the detachment area: freshly eroded channels, wet looking areas, and deposits from small debris flows – all perhaps signs of geothermally enhanced melt of snow and ice (Fig. 18). In the path of the ice–rock avalanche we find streamlined debris stripes similar to those found in the Kolka, Rasht 2019, and Aparejo avalanche paths (Fig. 18). As of late 2019, the glacier has not recovered from its 2007 detachment, and there are only a few small snow (or ice?) fields visible at its location. Elevation differences between 2000 (SRTM DEM, before detachment), 2007–2010 (ALOS PRISM, after detachment), and 2010–2015 (TanDEM-X; both the ALOS PRISM and TanDEM-X DEMs are multi-year composites, and therefore their date range is given) suggest that the detached glacier had average and maximum thicknesses of 21–28 and 50 m, respectively. Combined with the detached glacier area of about 0.5 km2, we estimate a detachment volume of roughly 10.5–14 × 106 m3 , which is in good agreement with Iribarren Anacona et al. (2015). The Tinguiririca case illustrates how glacier removal by the 1994 detachment was largely reversible as the glacier built up again to failure conditions under the climate at that time, whereas the 2007 glacier removal seems irreversible under the current climate and associated negative glacier mass balances in the region (Falaschi et al., 2018b; Braun et al., 2019; Dussaillant et al., 2019).

In air photos from 1962 and Corona-series reconnaissance-satellite images from 1967, we notice debris stripes in the valley similar to the ones visible after the 1994 and 2007 events and other cases described in this contribution. Over the glacier, the 1962 air photos clearly show that a glacier detachment similar to the 1994 event must have occurred not long (weeks, months, or a few years?) before the image was taken. Stripes of debris or ice remains after the avalanche are still well preserved in the 1962 images over the glacier (Fig. 18d). (The associated terrain sections are under snow cover in the 1967 Corona images.)

In the neighbouring valley to the east (Fig. 19) we note geomorphological traces that could be investigated further with regard to their origin from volcanic mass movements, earlier glacier stages, glacier surging, or more glacier-detachment-like events. In the lower part of this potential event path we find debris stripes similar to the ones from the Tinguiririca, Aparejo, and other avalanches of this study, largely unchanged since the first available Corona-series reconnaissance-satellite images in 1967 (Fig. 19c) and air photos from 1962. Only detailed field work would be able to rule out the possibility that these debris stripes consist of glacial flutes or small lateral moraines formed during a previous glacier extent or were produced by a glacier surge. It should also be investigated whether such debris stripes can remain largely intact after having been overrun by an ice–rock avalanche or if it is an indicator of the most recent event (see also Aparejo, where similar questions turned up).

In the upper part of this valley or in the potential event source area, we observe two noticeable changes over time. First, the tongue of this unnamed glacier (RGI60-17.01112, GLIMS ID: G289692E34781S) was much smaller in 1962 air photos than it is in 2020. Its advance of roughly 1.5 km since 1962, most of which occurred between 1975 and 1986 (Landsat), is in stark contrast to the pronounced shrinkage of the other glaciers in the area (Figs. 17, 19). Indeed, differencing a DEM, which we produced from 1962 stereo air photos, from the SRTM or TanDEM-X DEMs confirms elevation gains on the glacier tongue of up to 120 m between 1962 and 2000 (SRTM) or up to 150 m between 1962 and 2010–2015 (TanDEM-X; Fig. 19c). This development could point to the recovery of the glacier tongue after a removal some time before 1962. The volume gain of the glacier tongue between 1962 and 2000 is around 70 × 106 or 100 × 106 m3 between 1962 and 2015.

Second, we detected an elevation decrease in the glacier forefield of around 10–15 m between 1962 and 2000 which could point to the deflation of debris-covered ground ice deposited by a possible surge or glacier detachment (Fig. 19c). Third, a comparison of the 1962 air photos to contemporary images revealed a heavily bulged glacier surface between 3900 and 4650 m a.s.l. (Fig. 19a, b). Visually, this bulging is similar to the bulge found on Flat Creek Glacier prior to its 2013 detachment (Sect. 3.7). The slope of the glacier tongue is around 8–10∘ and around 35∘ for the steep upper part. The bulging upper part of the glacier is far above the regional permafrost limit so that polythermal ice conditions might well be found in parts of it.

We also examined a 1955 Hycon aerial photo (no stereo data to produce a DEM available to us), but visual interpretations from it remain inconclusive. Under illumination conditions that are very different from those of the 1962 images, the headwall glacier cover and the tongue seem larger in 1955 than in 1962, resembling rather its shape and extent of the 1980s. Below the position of the current (2020) glacier terminus, there seem to be dead-ice remains visible in the 1955 images. These are also visible in the 1962 images though they are shrunken. It remains thus to be clarified at this point to what extent the features observed can be explained by a surge or series of surges of the glacier before 1955 or between 1955 and 1962 or by a detachment-like event.

The conditions under which a detachment might occur can be qualitatively understood from a simple force balance analysis. To simplify the problem, we represent the potential detachment as a rectangular slab (Fig. 20a). The force balance between gravitational (left part of Eq. 1) and resistance forces (right part of Eq. 1) gives the following relationship: 1WLhρgsin⁡α=WLτb+2hLτd, and by solving for τd, 2τd=W2hρghsin⁡α-τb, where τd is the lateral shear stress (Pa), W is the slab width (m), L is the slab length (L is cancelling out in Eq. 1), h is the slab thickness (m), ρ is the density of ice (kg m-3), g is the gravitational acceleration (m s-2), α is the slab slope, and τb is the basal shear stress (Pa). Assuming that the glacier driving stress is originally in balance with the basal shear stress under normal conditions (τb=τ0), we have the following: 3τ0=ρghsin⁡(α), which gives the following equation by combining Eqs. (2) and (3) in a way that a ratio τbτ0 appears, which allows us to compare potential failure conditions to normal conditions: 4τd=Wρgsin⁡α21-τbτ0. A collapse can only happen if τd exceeds the mechanical resistance of the slab margins (critical shear strength τc). Using the work of Gilbert at al. (2018) in which the evolution of the force balance toward the collapse of Aru glaciers has been quantified, we estimate τc≅0.28 MPa for the Aru glaciers and apply this value to the simplified glacier slab. This allows us, using Eq. (4), to define a stability diagram as a function of detachment width and slope and the ratio τbτ0 (Fig. 20b).

This stability criterion seems to be respected by most of the detachments of our study (Fig. 20b), showing that the rough slab approximation may be a way to initially and qualitatively analyse which glaciers might be susceptible to detaching. In particular, the analysis shows which combinations of slope and width are unlikely to produce detachment when τd remains below τc even for τb=0 (i.e. complete loss of basal friction; green area in Fig. 20b). Under the provisional assumption that τc is similar for all the glaciers investigated, this analysis also shows that the detachments presented here happened for an average loss of friction between around 50 % and 100 % (i.e. 0<τb/τ0<0.5). In reality, it is reasonable to assume that the basal friction rarely goes all the way to zero. Additionally, the critical lateral shear stress will vary between glaciers, depending, for instance, on whether the critical lateral resistance is provided by ice or by morainic margins and their properties and on topographic forms of resistance like glacier curves or bedrock bumps. For the Amney Machen detachment, which did not follow our provisional criteria (it failed at a critical shear stress of 0.2 MPa, which is smaller than the 0.28 MPa estimated for Aru), the detachment flanks consisted of sediments and might thus be weaker than the ice margins along which the Aru glaciers detached. For some glaciers the effective width of the detachment is difficult to estimate as it varies along the glacier length, and we cannot be sure at which width the failure started to develop. For Tinguiririca, the effective detachment width is particularly uncertain as the glacier rested on a bed ramp rather than between valley or moraine flanks. It remains to be investigated how the deviation of the cross sections of the detached glaciers from the idealized slab influences our stability analysis.

As will be discussed in the following sections in more detail, several of the detachments presented here showed precursory accelerations, some clearly surge-like. In the present section we discuss to what extent the meltwater production associated with this precursory motion could feed back on the reduction of basal shear stress. We take the case of the Aru-1 event (based on data in Gilbert et al., 2018) as an example. The first major dissipative losses of this event were during the pre-detachment accelerated sliding, which attained about 0.5 m d-1. Whereas some energy drove crevasse development and brittle–ductile changes in ice crystals, we suppose that most of this phase of energy dissipation occurred at the bed by the grinding of rocks and ice and melting of ice. The typical glacier thickness was 100 m, so 1 m2 of bed area by 100 m column of ice at a density of 900 kg m-3 would have a mass of about 90 000 kg. Each day it slid downslope by about 0.5 m, including an elevation drop of about 0.1 m. The daily loss of potential energy of each such column of ice was thus about 88 300 J (about 1 W m-2). If the ice was at the melting point, which it was across much of the bed within the frozen perimeter (Gilbert et al., 2018), this energy could melt up to around 0.26 kg of ice per square metre of thawed bed per day. If this continued for 200 d at such rapid sliding, about 53 kg of liquid water would be generated per square metre of thawed bed area, amounting to a roughly 5 cm layer of water (some energy may also have been expended in crushing rocks and ice). This water could be expelled, or it could be contained within the frozen confines of the polythermal glacier. If confined, this water could be ingested into basal till, perhaps dilating the volume of till and spreading it over a larger fraction of the bed, thus reducing the total frictional resistance in the thawed parts of the glacier; the water itself could spread over a larger fraction of the bed, or it could pool against the frozen toe and margins of the glacier. Additional water – maybe a lot more – likely was provided to the subglacial bed environment from rainfall and snowmelt sources. The role of the above 5 cm average thickness of frictional meltwater could become important if (i) it is a substantial proportion of the thickness of the basal till, (ii) if it is a significant proportion of the meteorological meltwater that reaches the bed, and/or (iii) if it can spread laterally over a large part of the thawed bed.

Not all detachments presented here were preceded by longer phases of surge-like acceleration, and surge-like acceleration typically does not lead to glacier detachment. Still, the above estimates exhibit a feedback process that could through precursory acceleration facilitate a substantial reduction of basal shear stress and ultimately failure of soft beds.