Glacier detachments and rock-ice avalanches in the Petra Pervogo range, Tajikistan (1973–2019)

Glacier detachments are a rare, ::: but ::::::::: hazardous, : phenomenon of glacier instability, whereof only a handful have been documented to date. Common to all known cases are the large detached volumes of :: is ::: that many million cubic meters of ice and long runout distances. :::::::: detached :::: from ::: the :::: bed :: of :::::::: relatively :::::::: low-angle :::::: valley ::::::: glaciers ::: and :::::: turned :::: into :::::::::: long-runout :::: mass :::::: flows. Recently, two detachments of smaller size :::: such ::::::::::: detachments were observed in the Petra Pervogo range , north west of the Pamir mountains, : in : Tajikistan. Using a variety of satellite images, we ::::::: imagery, :::::::: including ::::::: Landsat :::: 1–8, ::::::::: Sentinel-2, :::::::: ASTER, 5 ::::::::: Tandem-X, :::::::::: Worldview, ::: and :::::::: Keyhole, ::: we :::::::::::: characterized :::: these :::::: events :::: and identified in total 9 detachmentsand several iceand rock avalancheswhich occurred :: 17 :::: mass ::::: flows :::::::: involving :::::: glacier ::: ice :::::::::::: (detachments, ::: ice, ::: and :::::::: rock-ice ::::::::: avalanches) :::: that :::::::: clustered in four different catchments between 1973 and 2019. The avalanche run out distances vary between :::::: runout :::::::: distances ::::: range :::: from 2 and : to : 19 kmand detached volumes range between 2 and : , ::: and ::: the :::::: largest :::::::: detached :::::: glacier ::::::: volume ::: was : 8.8× 10 m. Seven out of nine detachments :::::: Glacier ::::::: surging :::: seem :: to ::: be ::::::: frequent :: in ::: the :::: Petra ::::::: Pervogo :::::: range. :: 11 ::: out :: of ::: 13 ::::::::::: detachments, ::: ice :: or 10 ::::::: rock-ice ::::::::: avalanches occurred between July and September in years with temperature above the :::: mean :::::: annual ::: air ::::::::::: temperatures ::::: above ::: the :::: trend ::: of :: the : past 46-yearstrend. No active glacier surge was observed immediately before detachment, but elevation model (DEM) differences indicate a surge-like behavior about 10 years before the two largest detachments. Instead, one glacier retreated before detachment while the other remained stagnant before increased sliding pronounced the impending detachment. To put results into a regional context, we analyzed DEM differences over the entire Pamir range and found : . :::: The :::::::: relatively 15 :::: large ::::::: number :: of ::::::: locally :::::::: clustered ::::: events :::::::: indicates :::: that ::: the ::::: Petra :::::::: Pervogo ::::: range ::: has :::::::::: particularly ::::::::: favourable :::::::::: conditions ::: for ::::: glacier ::::::::::: instabilities. ::: The ::::::: images ::: and ::::::: geology :: of ::: the :::::: region ::::::: suggest ::: that ::::: easily :::::::: erodible ::::::::: lithologies ::: are :::::::::: widespread. ::::: These :::: soft :::::::: lithologies :::: may ::: be ::: one :::::: reason ::: for ::: the :::: high :::::: density :: of ::::::: surging ::::::: glaciers ::: that ::: we ::::::: detected :: in ::: the ::::: wider ::::: Pamir :::::: region : (237 surging glaciers, predominantly in the north-western part where soft and fine-grained rock-types are common. We are confident that no major events were missed due to lack of satellite data, because destroyed vegetation remains visible in the normalized difference 20 vegetation index (NDVI), several years after large mass flows, e. g. about 10 years for the Kolka-Karmadon rock-ice avalanche. From the large number of detachments which occurred under very similar conditions we conclude that rising temperatures ::::: total). ::: We ::::::: conclude :::: that :::: high ::::::::::: temperatures, : combined with soft, fine-grained sedimentsare very critical components favouring : , :::: may ::::::: increase ::: the ::::::::: likelihood :: of ::::: mass ::::::: wasting ::::: events :::: and :::::: appear :: to ::: be :::::: critical :::::: factors ::::::::: facilitating : the detachment of entire valley


Introduction
Glacier detachments are extremely rare events but the scientific understanding of these events is rapidly evolving. They occur when large volumes of glacier ice detach from valley glaciers with relatively low surface slopes (10°to 20°) and turn into highly mobile, ice-rich mass flows. Evans and Delaney (2015) list glacier detachments, together with ice avalanches, as one of three classes of catastrophic mass flows in glacierized mountain environments that are pertinent to this work. The classes 20 are distinguished by their starting mechanism and the involved material. Both glacier detachments and ice avalanches mainly involve glacier ice, but ice avalanches are much more frequent and typically originate from steep (hanging) glaciers. Rock avalanches -with sometimes long runouts if they descend onto glaciers or snow covered terrain -form a second class; the combination of the first two classes, or mass movements that involve both ice and rock (Evans and Delaney, 2015), are classified as ice-rock or rock-ice avalanches. For all three classes, potential energy is transformed into kinematic energy and 25 into frictional heat. Frictional heating, and sometimes entrained sediments (Moore, 2014, Sect. 5.2.2), increase the liquid water content which can enhance the mobility of the resulting mass flows (Schneider et al., 2011;Evans and Delaney, 2015;Davies, 1982). The high mobility leads to much longer runout distances compared to pure rock avalanches (Schneider et al., 2011), and in turn increases the potential for damage to inhabited areas (Petrakov et al., 2008).
Several past events -including the 2002 Kolka-Karmadon rock-ice avalanche (Drobyshev, 2006;Huggel et al., 2005;Evans 30 et al., 2009), the 2016 Aru Co twin glacier collapse Gilbert et al., 2018), the 2013 and 2015 Flat Creek detachments (Jacquemart et al., 2020;Jacquemart and Loso, 2019), as well as comparable events reported from China and Argentina (Paul, 2019;Falaschi et al., 2019), are well described by the definition of glacier detachments offered by Evans and Delaney (2015) because they involved "the decoupling of a glacier ice mass from its bed and catastrophic detachment of a large volume of a valley glacier". We therefore adopt this term when documenting the newly discovered detachments, as well 35 as when referring to events described elsewhere (e.g., Kolka-Karmadon detachment, Aru detachments).
The reasons for glacier detachments are not yet fully understood, but several factors seem to play a major role: Water has been found to be the main cause for the drastic reduction of basal friction that is key for a glacier detachment Gilbert et al., 2018;Jacquemart et al., 2020), but stress changes due to loading from rock or rock-ice avalanches on the glaciers have also been invoked as possible triggers (Evans et al., 2009;Kääb et al., 2020). Fine grained sediments or 40 weak bedrock underlying the glaciers have been found for all glacier detachments, presumably facilitating the storage of large amounts of water leading to the necessary loss of friction Gilbert et al., 2018;Jacquemart and Loso, 2019).
Also, ice-sediment mixtures have been shown to experience profound weakening at temperatures close to the melting point (Moore, 2014). In many cases, a close proximity to surging glaciers has been documented; in some cases, the detached glaciers themselves exhibited a surge-like behaviour before the detachment , or had a prior history of surging. satellite images, including the entire Landsat archive. We subsequently used this inventory to put the glacier detachments in 60 context with the local geology, climate conditions, the regional distribution of surge-type glaciers, and pre-detachment glacier dynamics.Finally, we also put these findings in context with glacier detachments known from elsewhere, in particular the well described events at Kolka, Aru and Flat Creek.

Study site
The Petra Pervogo range (also called Peter the First or Peter the Great range) is situated in central Tajikistan, north-west of the 65 Pamir mountain system. It extends east to west for about 200 km between the Surkhob river to the north and the Obikhingou river to the south, both of which drain into the Vaksh river at the western end of the range. In the West Petra Pervogo range, shown in Fig. 1, we identified four catchments which showed repeated large mass flows. Two detachments, which happened in 2016 and 2017, were mentioned in (Dokukin et al., 2019) and on Twitter (Dokukin, 2018). A third detachment, which happened in 2019, was found during this study and independently by Kääb (2020). In the catchment of the Degilmoni Poyon river (DP in Fig. 1) we identified a glacier detachment which occurred in 2019 (abbreviated as dp-19). It resulted in a debris flow which almost reached the village Degilmoni Poyon, located 9 km downstream.
The glacier detached between 2860 and 3360 m asl (above sea level) at about 38.988°N, 70.694°E.
In the catchment of the Shuraki Kapali river (SK in in Fig. 1), 13 km upstream of the village of Tojikobod (Tadzhikabad,   75 1588 m a.s.l.), a series of detachments and ice avalanches occurred between 1973 and 2019. The nearby villages Kapali and Fathobod experienced some infrastructure damage from an event on 28 August 2016. The largest detachment from this catchment occurred in 2017 (abbreviated as sk-17 in the following), when ice masses detached from between 3300 and 4000 m asl at about 38.974°N, 70.844°E.
In the catchment of the Shikorchi river (Shi in Fig. 1), we identified a series of mass flows between the years 2000 and 2017, 80 most of them rock-ice avalanches originating at elevations between 3000 and 4000 m asl (39.026°N, 70.933°E).
In a side valley of the Shaklysu river (Sha in in Fig. 1

85
The western Petra Pervogo range is composed mainly of Cretaseous-Neogene sedimentary rocks. The catchments DP and SK are made up of redstone, aleurolite, claystone, conglomerates and limestone. Striking erosional features and thick glacial debris cover support the fact that soft lithologies are widespread. The Petra Pervogo range is located south of the Vakhsh thrust system where shallow earthquakes in the upper 15 km of the crust are frequent (Schurr et al., 2014). Similar sedimentary rocks are prolific in the north-western corner of the Pamir mountains (lime-, clay-, and sandstones, as well as conglomerates, aleurolite, 90 gypsum and marl; see geological map by Ibrohim et al. (around 1974)). The prevalence of such rocks, which may be easily erodible by glaciers and freeze-thaw processes, is spatially correlated to the particularly high density of surging glaciers present in the area (Goerlich et al., 2020).

Climate
Two meteorological stations, one located at Rasht/Garm (1316 m, 39.02°N, 70.37°E), 40 km west of SK in the Surkhob valley, 95 and the other located above the Obikhingou river at Lyairun (2008 m, 38.89°N, 70.93°E), 12 km southeast of SK, indicate a mean annual precipitation of 700-1000 mm yr −1 (Williams and Konovalov, 2008) and a mean annual air temperature (MAAT) of 10.7°C and 7.1°C, respectively, resulting in a temperature-lapse rate of -0.52°C per 100 m. The zero-degree isoline in the region is therefore at around 3300 m, and a global permafrost map indicates that permafrost is patchy (Obu et al., 2019).
Vegetation grows until about 3500 m and glacier tongues reach down to 2700-3200 m. Based on Sentinel-1 radar backscatter 100 data we determined that snow melt at ∼ 4000 m starts around mid April every year, and melting temperatures last until October. A temperature increase of 0.42°C over the last 40 years has been observed for the Pamir mountains, with an increase of almost 1°C in fall and winter (Finaev et al., 2016).

Data and methods
Very little in situ data was available to us, so this study is primarily based on remote sensing imagery. We combined optical 105 and radar images, as well digital elevation models (DEMs) to identify, map, and characterize mass flows in the Petra Pervogo range as well as the distribution of surging glaciers in the range and the larger Pamir region.

Detection and classification of mass flows
We analyzed the entire Landsat (L1-L8), Sentinel-2 (S2) and ASTER archives, as well as all freely available reconnaissance Keyhole (KH3, KH-4A/4B, KH-9) images to identify, classify, and characterize large mass flow events in the glaciated envi-110 ronment of the western Petra Pervogo range (Fig. 1). A temporal overview of analyzed acquisitions is shown in Fig. 2.
For event detection, we searched for the abrupt disappearance of glaciers and also for the appearances of bright (ice rich) and dark (sediment rich) deposits in the valleys (for examples see Appendix). In addition, we looked for removal of vegetation, and changes in surface color indicating overtopping of landscape by debris flows. To detect such changes, we visually compared images from consecutive years but acquired during similar snow conditions at the same month (or day, if available) of each year.

115
For that we mainly analyzed imagery between July and Sept where snow and cloud cover was minimal and where vegetation showed a strong near-infrared (NIR) signal. In addition, we also compared consecutive images to detect events that occurred in winter or that did not leave traces visible to be detected in the next summer. We chose spectral bands by the best available resolution (Table 1), followed by their ability to discriminate vegetation, ice, rock and wet sediments. Where possible, we chose longer wavelengths which better penetrate aerosols. We used the moisture-sensitive short-wave infrared channel (SWIR2) to 120 distinguish wet and dry sediments (Kääb et al., 2014;USGS, 2020). To identify vegetation cover we used the NIR channel; For L7 we used the panchromatic (PAN) channel (L7) which covers the NIR spectrum, but provides a higher spatial resolution.
For L8 we averaged the higher resolution panchromatic band 8 with the vegetation-sensitive NIR band 5. To identify snow and ice we used the red (R) or green (G) channel from the visible spectrum. To narrow down the date of events we also analyzed selected optical Planet imagery and Sentinel-1 (S1) radar imagery. We analyzed all images at a scale of approximately 1:25'000.

125
For a more detailed analysis of detected events we zoomed in. We considered only events where horizontal length from release area to deposit end exceeded about 2 km. Due to the growing availability of imagery (see Fig. 2) our collected dataset is very likely biased towards more recent years. As glacier detachments are extremely rare events, we aimed for detection of as many as possibly events, rather than on temporal consistency of the dataset as done by others (e.g. Bessette-Kirton and Coe, 2020).
Determining the nature of the detected events is rarely a straight forward task, but we try to offer our best assessments based 130 on the following criteria: -Glacier detachment (d): A glacier was visible prior to the event and lies within a GLIMS inventory (Glacier Land Ice Measurements from Space) polygon; the glaciers were located at the bottom of a valley or in topographical depressions; exposed bedrock or shadows indicated the removal of large amounts of ice; downstream deposits consisted of mostly ice.

135
-Ice avalanche (i): Only small amounts of ice were removed and the glacier seemed mostly intact; downstream deposits consisted of mostly ice. -Rock-ice avalanche (r/i): Release area included likely some ice but also rock; deposits were mostly ice free.
-Rock avalanche (r): Release area included mainly rock; deposits were mostly ice free.

Mass flow descriptions 140
For each detected event, we determined the release area and the slope of the release zone. To characterize the mobility of the mass flows, we determined the angle of reach α, calculated by tan α = H/L from the horizontal path length L and the total fall height H measured from the top of the release area to the lowest point of the runout. The angle of reach (or mobility index or Fahrböschung) corresponds to the average friction coefficient of the mass flow (Scheidegger, 1973). We also measured the total impact area and the maximum height of the flows' trim lines using the elevation information of the SRTM DEM embedded in 145 Google Earth Pro.
Precise estimation of volume changes requires the availability of timely elevation models before and after an event. Unfortunately, this was only the case for the event dp-19, for which three pairs of World View stereo images (Neigh et al., 2013) could be processed into DEMs using the SETSM algorithm (Surface Extraction with TIN-based Search-space Minimization from Noh and Howat (2017)). We coregistered the DEMs to each other following Nuth and Kääb (2011). For all other detected 150 events, the available DEMs (Table 2) had either no precise time stamp or the detected events happened several years before or after the DEM acquisition, inhibiting precise volume estimates. In some cases, however, the DEM time series provided insight into a glacier's dynamics prior to the events.
To estimate the uncertainties of the volume estimates we masked all areas impacted by the events and tiled the DEMs into n 2 tiles (n ranging from 2 to 200). By calculating the median height change (dH) per tile and relating this to tile size, we 155 get estimates of the average per-area dH error (Miles et al., 2018). This empirical error metric accounts for all error sources, including differences in snow cover, processing errors etc. In the World View images we masked obvious clouds (large areas with a DEM difference beyond ±130 m) in addition to the area impacted by the glacier detachment.

Pre-event glacier dynamics
Several studies have described surge-like behavior, increasing flow velocities, or opening crevasses prior to glacier detachments 160 and ice avalanches Jacquemart et al., 2020;Faillettaz et al., 2011). Where the data permitted, we tried to detect and describe such behavior.
We used high resolution S2 and L8 imagery to measure flow velocities and crevasse opening prior to the events sk-17 and dp-19. Velocities were determined by manual tracking of surface features and by measuring the width of the opening rupture lines.
We also tracked any surging or surge like mass redistribution using DEMs from SRTM, TanDEM-X (TDM), the ALOS World 165 DEM 3D (W3D), and World View (WV) stereo imagery (Farr et al., 2007;Krieger et al., 2007;Tadono et al., 2016;Neigh et al., 2013;Noh and Howat, 2017). We analyzed six TDM pairs acquired between 03 May 2011 and 05 September 2014, and generated DEMs using the InSAR processing algorithm detailed in (Leinss and Bernhard, 2021) to derive the surface dynamics from DEM differences.

Regional surge patterns 170
To compare the geometric characteristics of detected glacier instabilities within a wider regional context, we mapped glacier surges in the entire Pamir mountains, that occurred between 2000 to 2011 by differencing the C-band SRTM and the optical W3D, both at 30 m resolution, horizontally aligned following Nuth and Kääb (2011). We analyzed DEMs from 37-39 North and 67-75 East. For the SRTM DEM an absolute vertical accuracy of 6 m is given in (Farr et al., 2007) but the C-Band radar can penetrate up to 10m into dry snow and firn (Rignot et al., 2001). For the W3D a vertical accuracy of 5 m is given (Tadono For mapping of glacier surges, we considered glaciers as beeing an active surge phase when the glacier showed a surface height increase of more than 10 m over the glacier tongue accompanied by surface lowering further upstream. We consider glaciers being in a quiescent surge phase when surface lowering over the glacier tongue exceeded 10 m and a significant 180 surface height increase was visible upstream, in a possible reservoir area. To determine the slope of the surging part of a glacier we measured the horizontal length and the elevation difference of the area that showed the surge-like (wave-like) elevation change pattern.

Meteorological and seismic data
To analyze climatic influences on glacier detachments and ice/rock-ice avalanches, we used data from the two meteorological between the Lyairun and ERA-Land temperature and shifted the temperature data of the Lyairun station by +7.6°C to match Table 3. Type and characteristics of detected mass flows determined as described in Sect. 3.1 and 3.2. Empty fields indicate quantities that could not be determined. Surge-like instabilities were observed several years before the sk-17 and dp-19 events ("yes" in paranthesis) but not immediately before the detachment. The sk-16b event transformed into a debris flow, possibly after entraining material left by the sk-16a event.
To assess earthquakes as triggering factors, we used data of seismic events that occurred within a range of about 100 km around the Petra Pervogo range. We selected the earthquakes which occurred within the time period given by a pre-event satellite image and a post-event image. To capture delayed triggering by earthquakes we also selected earth quakes up to two days before acquisition of the pre-event image. Then we assessed the earthquakes's magnitude and the distance to the catch-195 ments where mass flows were detected and compared the earthquake's distance and magnitude to the threshold for triggering of disrupted landslides according to (Jibson, 2013).

NDVI for mass flow recognition and vegetation recovery analysis
The older, available imagery showed gaps of a few years in which mass wasting events could have happened without being noticed. However, events with long runouts may remove or bury vegetation which can take years to recover. To assess vegetation 200 recovery times, and to estimate how likely large events might have been unnoticed in post-event imagery containing a vegetation sensitive channel, we analyzed time series of the NDVI = (NIR − R)(NIR + R) from the band combinations (B5, B4) and (B8, B4) for LS8 and S2, respectively, of the two recent detachments, dp-19, and sk-17. We compare the results with vegetation recovery in the runout of the Kolka-Karmadon glacier detachment.  4 Results

205
Our analyses revealed a very high activity of mass wasting events in the western Petra Pervogo range. In particular, we have detected two large-volume glacier detachments, as well as several smaller glacier detachments, ice avalanches and rock-ice avalanches. Table 3 summarizes the characteristics of all detected events; Table 4 lists satellite images used to narrow down their date of occurrence. The following two sections describe the main detachment events, followed by short descriptions of all other events grouped by (sub)catchments.     (Fig. 4b,c). In a Google Earth image from 30.07.2007 the glacier appears heavily crevassed (Fig. 4d), indicating another active phase of advance which last until at least 2008 according to L7 imagery. After that, the glacier entered a pre-detachment quiescent phase. In L7, L8, and S2 data that glacier appears progressively sediment covered and no special activity was detected

Other events 4.3.1 Shuraki Kapali (SK) catchment
The Shuraki Kapali catchment appears to be a hotspot for glacier detachments and ice or rock-ice avalanches. Distributed across three small sub-catchments, the GLIMS database lists five small glaciers in the upper part of this drainage (Fig. 7). We 255 briefly describe the detected events, grouped into their respective sub-catchments, from west to east.
In the western part of the Shuraki Kapali catchment the GLIMS data base lists two small glaciers from which at least five mass flows originated. Extensive debris cover on the two glaciers made a precise delineation of the detached areas and unambiguous classification of the events difficult.
-In July 1994 the lower part of a glacier with the GLIMS ID G070839E38975N broke away (sk-94) and resulted in an 260 rock-ice avalanche with an approximate runout distance of 2.7 km. We did not find earlier events in this catchment but KH imagery indicate strong erosion and sediments below the glacier (Fig. A3a).
-In September 2004 a slightly larger part detached from the same glacier (sk-04) and resulted in a mass flow with an approximate runout distance of 2.9 km. The detachment zone and the avalanche are visible in Fig. A3c. Additional ice fell off from the upper scarp of the detachment zone a few days later (arrow in the inset of Fig. A3c) resulting in a similar 265 runout distance of 2.5 km.
-In early September 2010 ice continued to break off from the remaining parts of the glacier and resulted in a rock-ice avalanche (sk-10).
-For 28 August 2016, local media reported a mud-flow as a result of glacier break off (Tajik telegraph agency, 2016; Radio Ozodi, 2016). We determined a glacier area of 160 × 10 3 m 2 , indicated as sk-16b in Fig. 7, which detached, where ten buildings and a bridge were damaged or destroyed and several cattle were swept away. The mud-flow reached the Surkhob River at 1507 m of altitude (inset in Fig. A4b), still containing pieces of ice according to photographs in media, and blocked temporarily the Shuraki Kapali river (Radio Ozodi, 2016).
-Between 21 and 23 June 2019 a rock-ice avalanche (sk-19) was released at the same place as sk-16b. However, Google Earth imagery indicates that a deeper layer of rock or ice has detached. The event was followed by a minor ice avalanches In the central part of the Shuraki Kapali catchment the GLIPS data base lists a glacier with the ID G070846E38972N. Here, we identified three events, two glacier detachments, one followed by an ice avalanche.  Fig. 6d. The detachment area of 170 × 10 3 m 2 was derived from L7 imagery one year after detachment (Fig. A5a). In the detachment area, DEM difference between the SRTM and the W3D showed a height loss of up to 40 m (15 m on average; red area in Fig. 6). From DEM differences we estimate a volume loss of 290 at least 2.9 ± 0.3 × 10 6 m 3 for sk-03. The volume is very likely larger because the W3D is mainly composed from data acquired several years after the event, between 2006 and 2011.
-In late August 2006 glacier ice with an area of 55 × 10 3 m 2 (sk-06) was release just above the detachment scarp of the sk-03 event, resulting in a runout of 3.4 km, Fig. A5b. The likely rupture line of this event, and the detached area of sk-03 below, is visible in a Google Earth image from 13 August 2008 (Fig. 6c).
The resulting mass flow travelled 5.6 km over a height loss of 1200 m, corresponding to an angle of reach of 12.4°. TDM imagery and DEM differences indicate that the valley exposed by the sk-03 event has partially filled up with ice.
In the eastern part of the Shuraki Kapali catchment a KH-09 reconnaissance image from 03 August 1973, indicates a rockice avalanche (sk-73; Fig. A2b), likely originating from the the glacier that produced the sk-17 detachment. From the runout

305
In the catchment of the Shikorchi river, we identified a series of large mass flows which travelled over steep glaciers but we could not determine how much ice was entrained during flow or involved in the release area. The runout did not show clear traces of ice, therefore we classified them as rock avalanches. This classification is supported by the relatively steep slopes (26-38°) and the low mobility (angles of reach 17.6-19.6°) listed in Table 3.

Shaklysu catchment (Sha)
In a side-valley of the Shaklysu river a very small glacier with the GLIMS ID G070995E39014N is located. -In August 2006, satellite imagery indicate a similar event (insets in Fig. 8).

Meteorology and seismic activity
Almost all detected events (14 out of 17) occurred in years where the mean annual air temperature (MAAT) was above the 46 years trend (Fig. 9). Only the sk-94 and sk-06 event and the shi-09 rock avalanche occurred in years with a MAAT below the trend. Except for rock avalanches, all events happened in between June and September which are the warmest months of the year. We interpret this in the sense that temperature has a very strong impact on the occurrence of glacier detachments. No 330 correlation to precipitation was found.
The magnitude and distance of all earthquakes which occurred within a radius of 500 km of the SK catchment are shown in Fig. 10 as gray dots. The solid lines indicates the threshold for triggering disrupted landslides (Jibson, 2013). As we do not know the sensitivity of glacier detachments and rock-ice avalanches to earthquakes, we shifted the threshold disrupted landslides by one and two earthquake magnitudes (dashed and dotted line). Earthquakes that occurred between the pre-event 335 and post-event satellite image (Table 4) and that are close enough and strong enough to be at least below the dotted line (magnitude for disrupted landslides -2) are shown as black bullets. We found no mass flow events that could have been triggered by earthquakes below the dashed line (magnitude for disrupted landslides -1). Because stronger earth quakes did not trigger any mass flows, we conclude that rock/ice avalanches and detachments are not especially sensitive to earth quakes. The solid lines indicates the threshold for triggering disrupted landslides (Jibson, 2013); the dashed and dotted lines represent the same line but shifted by one and two magnitudes, respectively. Black dots indicate earth quakes that occurred between an pre-and post event image of the analyzed events and that are closer or stronger to be located below the dotted threshold.

340
In total, we identified 237 glaciers in the entire Pamir mountains where DEM differences (W3D -SRTM) showed a height change indicating either an active surge or a quiescence phase. Of these 188 showed both an elevation increase at the terminus and a decrease further up, 32 glaciers showed only an elevation increase at the terminus and 17 seemed to be in a quiescent phase with strong melt at the tongue but mass gain in a possible reservoir area.
In the Petra Pervogo range we found four surge-type glaciers, listed from East to West: at 38.925°N, 70.524°E a glacier The comparison in Fig. 11 of the slope and length of all surging glaciers with the detached glaciers of the the largest events, 350 dp-19, sk-17, sk-03 and sk-04, and in addition with the Aru-and Kolka-Karmadon detachments, shows that glacier detachments occur predominantly for short but steep glaciers, at least when compared to glaciers which showed a surge-like instability in the past.

Retroactive avalanche detection using NDVI
The largest avalanches in this study, sk16b, sk-17, sk-19, and dp-19, were identified in satellite imagery by destruction of 355 vegetation in the valleys. The analysis of time-series of the NDVI evolution in the DP-and SK-catchment, Fig. 12, shows that vegetation did not recover within the two years of the events. In the runout zone of the Kolka-Karmadon detachment, where a suitable long satellite time series exist and where no repeated avalanches occurred, vegetation recovery to pre-detachment NDVI values took around 10 years (Fig. A1). The vegetation covered runout zones of the SK/DP catchment at roughly 2500 m and the runout of the Kolka-Karmadon detachment at 1800 m (Haeberli et al., 2004) show a similar climate: for SK/DP we 360 obtain a MAAT of +4.5°C which is comparable to the a MAAT of +4.0°C below the Kolka-detachment. The similar climatic conditions indicate that vegetation recovery times are comparable. Therefore, we conclude that the chance of missing long runouts of mass flows that reach vegetated areas is very low when imagery every few years is available.
Unfortunately, most other avalanches travelled in already eroded valleys, therefore it was difficult to detect them by means of vegetation change only. We observed that the white color of ice avalanches quickly disappeared within a few days. Therefore, 365 it is likely that smaller events have been missed, especially in years with frequent cloud cover.

Discussion
The numerous recent discoveries of glacier detachments around the world Gilbert et al., 2018;Falaschi et al., 2019;Paul, 2019;Jacquemart et al., 2020) have raised important questions about the conditions and triggers leading to these events. Our analysis of the 46-year of satellite record over the Petra Pervogo range has revealed a cluster of such events in a small geographical area that provides additional understanding of these catastrophic events, in particular with regard to the link between surging glaciers and glacier detachments, and the influence of temperature and seismic activity.

Detachment detection
Analyzing the entire satellite record is not a fool proof approach, since clouds and shadows can hamper the detection of certain events, but we always compared multiple consecutive images as well as images acquired in the same month of consecutive 375 years. While the traces left by smaller events easily disappear against the background of loose sediment and hillslopes free of vegetation, large events that reach vegetated areas leave distinct traces that can be detected for several years. Our analysis of vegetation recovery at Kolka-Karmadon (approximately 10 years), and the fact that we discovered sk-17 and dp-19 in this fashion, demonstrate how the NDVI and the vegetation sensitive NIR channel are good means to detect long-runout events in remote sensing imagery, even years after they happened. Closer to the source, where there is typically no vegetation, the 380 moisture sensitive channels SWIR1 and SWIR2 of Landsat -7 and -8 allow for the detection of sediment-covered ice, until at least a few weeks after detachment. Lastly, the low resolution of 80 and 30 m of Landsat 1-5, which lack a higher resolution panchromatic channel and especially the lower number of available images, could impede the detection of some early events. To complement the drawbacks of detecting optically visible changes differencing high resolution DEMs, acquired within a period of months to a few years, is undoubtedly the most reliable way to detect drastic changes in glaciated catchments; however, such 385 DEM data is currently not acquired operationally and is only sparsely available in time and coverage. We found that weatherinsensitive radar imagery is helpful to detect abrupt changes, but the bright backscatter signatures of ice avalanches disappears within a few days due to melt. Due to increasing availability of imagery (Fig. 2), we are relatively certain that our dataset is biased towards more frequent events, hence, no conclusion can be drawn from the relative frequency of detected events.
In contrast to the detection of past events, detection of glaciers that may be prone to detach in the future is a much more 390 difficult task. On sk-17 and dp-19, increased crevassing could be only seen in high resolution images a few weeks prior to the detachment. This makes it extremely difficult to identify possible instabilities sufficiently early, especially when a glacier is not inspected on a regular basis. Similarly, the Aru glaciers also showed increased crevassing just a few weeks before their detachments . Indeed, even the supposedly tell-tale crevasses don't always reliably predict a detachment.
For example, a small glacier near the Gulyia-Ice cap in the western Kunlun Shan has been showing detachment-like crevasses 395 since early 2018 (Leinss et al., 2019), but has remained stable so far, likely due to the stabilizing effect of its very broad tongue.
Automated near real-time velocity monitoring using very high resolution sensors could be another option for early glacier hazard identification. However, based on our experience, the detached glaciers in the Petra Pervogo range are too small for current optical or radar sensors to provide reliable velocity estimates. Increased data bandwidth and imaging capabilities of future sensors and high-repeat rate DEM differencing satellites could provide the required data for early detection of possible 400 detachments. In the specific catchments of this study, where large mass flows occur frequently, in situ observations by radar or cameras could very likely act as an relatively cheap warning systems to inform local population in time.

Detachment characteristics and triggers
Fundamentally, the question of which events to classify as glacier detachments -failures of low-angle valley glaciers that involve substantial amounts of the glacier -is a tricky one when the observations are purely based on remotely sensed imagery.

405
In our study region, the task is further complicated by wide spread debris cover, which makes it hard to delineate glaciers.
While the boundaries of the glacier detachment category are certainly fuzzy, we have classified four of the 16 detected events listed in Table 3 as glacier detachments. The posterior analysis of the detachment events shows that all share the characteristic low to medium surface slope of the detached area (15-20°) and that all occurred in a location where the GLIMS database (Raup et al., 2007) indicated the presence of a glacier. When using only satellite imagery for classification, the transition from 410 detachment to rock-ice avalanche seems to be continuous as the amount of detached rock is hard to quantify and deposits can contain entrained sediments or sediments from the bedrock. Some of the events classified by us as rock-ice avalanche might well be glacier detachments of glaciers with a relatively steep slope (20-25°). Remarkably, all events presented in this study happened within a roughly 30 km radius and the glaciers in the catchment areas present very similar characteristics regarding elevation and aspect (Table 5), with the SK catchment, for which the GLIMS data base lists five separate glaciers, appearing to 415 provide particularly favorable conditions for detachments and rock-ice avalanches.
Henceforth, we focus our discussion on these events, in particular on the largest detachments sk-17 and dp-19. In comparing these two events with other detachments described in literature (in particular Aru, Kolka-Karmadon and Flat Creek), we find similarities in slope, lithology and the time of year of the events. Both images and the described lithology (sedimentary) suggest that the easily erodible bedrock and soft sediments are abundant in our study area. Similar to Kolka glacier, dp-19 was below 420 a steep headwall and detached at the Bergschrund, so that the resulting mass movement involved basically the entire glacier.
As has been reported for other glacier detachments Gilbert et al., 2018;Jacquemart and Loso, 2019), there is a remarkable proximity, or in some cases overlap, between detaching and surging glaciers. Like others, we identified hundreds of surging glaciers throughout the Pamir, and the spatial distribution of the surging glaciers identified in our study is similar to Goerlich et al. (2020, Fig . 6). By comparison of the spatial distribution of surging glaciers with the rock types according to 425 the geological map by Ibrohim et al. (around 1974) we found that surging glaciers occur predominantly in regions with soft and fine-grained rock-types. It is noteworthy, though the importance and effect not yet well understood, that the glaciers that later detached (sk-17 and dp-19 in our study, but also the Aru and Kolka glaciers) exhibited a slightly steeper slope and were relatively short compared to their non-detaching surging neighbors (Fig. 11). Both sk-17 and dp-19 have surged in the past, but neither were in the midst of a surge immediately before their detachment, nor did they show any surge-like behavior. They 430 did, however, show a significant acceleration in the weeks prior to the detachment. Therefore, we do not believe that sk-17 or dp-19 were the consequence of a "runaway surge", but that both glacier surging and glacier detachments are favoured by a soft sedimentary bedrock. We rather conclude that the detachments were triggered by external drivers: because velocities increased during or after snowmelt, we suspect that increased liquid water input played a crucial role in lubricating the glacier base or saturating the underlying glacier bed . This idea is supported by the fact that all detachments in this study 435 happened in summer (end June-September), when more liquid water is available making it's influence on the glacier dynamics greater. We did not find any indication that earthquakes could have triggered the detachments or rock-ice avalanches. Instead, we have observed that 14 out of 17 mass movements, including 11 out of 13 detachments, ice or rock-ice avalanches occurred in years when the mean annual air temperature was above the linear trend of the past 46 years. Even though we think that our dataset is biased towards detection of more recent events, the comparison to the linear trend provides an indicator for the 440 sensitivity to temperature, while the comparison to the average temperature should results in a observational bias that we tried to avoid.
The fact that relatively short and steep glaciers (compared to their surging neighbours) show detachments could be related to the reason that short glaciers are more likely to have a more homogeneous slope compared to long glacier. When enhanced melt water lubricates the homogeneous base of a short glacier it is more likely to detach compared to a long glacier where 445 lubrication might cause a more local effect and could possibly init a surge-cycle when a sufficiently high mass imbalance is present.
All of the investigated events were very mobile, though at first glace, their mobility, characterized by an angle of reach of around α = 10 − 15°, was lower than that of the events at Aru and Kolka (α = 5 − 8°) (Huggel et al., 2005;Kääb et al., 2018).
The lower mobility can be partly explained by the smaller volume involved ( Petra-Pervogo: 3 − −9 × 10 6 m 3 , the others 70-450 130×10 6 m 3 ). However, if we compute the ratio V /L between detachment volume and runout distance, the ratio is one to two orders of magnitude smaller compared to the Kolka and Aru detachments, indicating an extremely high mobility. This could be a consequence of the path geometry, which channelized the avalanches over a very long distance in a small area. The valleys of easily erodible sediments provided few obstacles and thus small energy loss. In addition, we think the exceptionally long runout of 19.1 km of the event sk-16b, which angle of reach of α = 7.7°is comparable to the other large events, is caused by 455 entrainment of the ice-water-sediment mixture deposited in the catchment by the sk-16a event five weeks before. A video of the event shows that the debris flow is almost as liquid as water (Radio Ozodi, 2016).

Conclusions
In this study we built an inventory of glacier detachments and ice or rock-ice avalanches which occurred in the western Petra Pervogo range in Tajikistan. Compared to a handful of other large glacier detachments around the entire world we found a 460 cluster of at least four relatively small detachments and seven rock-ice avalanches within a radius of 30 km. The fact that multiple detachments occurred under very similar conditions (elevation, aspect, size, meteorological conditions) allows for studying external driving factors which can trigger the detachment of a valley glacier. We found that detachments occur in summer and in years with annual mean air temperature above the 46-year trend, indicating that high temperatures are an important factor favouring glacier detachments and rock-ice avalanches. The comparison to the temperature trend instead 465 to the mean temperature reduces the observational bias resulting from the increased availability and resolution of satellite imagery. Despite being a seismic active region, we found that earthquakes are very unlikely to be the cause of mass wasting events in our study site. Similar to other detachments, the glaciers in our study rest on a bedrock of soft sediments. For a rockice avalanche end of August 2016, we think that the entrainment of sediment-ice debris mixture from a previous detachment of relatively small volume five weeks before was the reason of the resulting, extraordinary long mud flow of 19.1 km. We 470 also observed a spatial correlation between the occurrence of surging glaciers in the Pamir mountains and soft, fine-grained sediments. However, we did not observe that the studied glacier detachments were a consequence of surging but we think that soft sediments are a prerequisite for detachments and at least a favouring factor for hydrologically controlled glacier surging. From the fact that the studied detached glaciers are shorter and steeper compared to surging glaciers in the same region we hypothesize that melt water penetrating to the glacier base can lubricate major parts of the relatively small bedrock 475 of soft sediments which then can lead to detachment of the entire glacier, especially if the glacier is relatively steep and the destabilized area is not supported by a stabilizing tongue of lower slope. In contrast, for longer glaciers it is unlikely that the entire glacier loses friction at the bedrock and it might instead be more likely that the glacier shows a temporary surge-like advance.
Code and data availability. S2, L1-8, ASTER, and Sentinel-1 data are available in the Google Earth Engine data catalogue and were pro-480 cessed with the Google Earth Engine (Gorelick et al., 2017) with Java scripts available on request from the authors. Some Copernicus S-2 data and USGS L8 data were also processed by ESA and downloaded form the Sentinel hub with the EO Browser: https://www.sentinel-hub.
TanDEM-X data is available from DLR https://tandemx-science.dlr.de/ and was provided by the proposal leinss_ XTI_ GLAC6600. Digital-Globe data were provided by the Commercial Archive Data for NASA investigators (cad4nasa.gsfc.nasa.gov) under the National Geospatial-

485
Intelligence Agency's NextView license agreement. The SRTM DEM is available from the USGS; The W3D is commercially available at