remote sensing

Article The 2015 Surge of Hispar Glacier in the

Frank Paul 1,*, Tazio Strozzi 2, Thomas Schellenberger 3 and Andreas Kääb 3

1 Department of Geography, University of Zurich, 8057 Zurich, Switzerland 2 Gamma Remote Sensing, 3073 Gümligen, Switzerland; [email protected] 3 Department of Geosciences, University of Oslo, 0316 Oslo, Norway; [email protected] (T.S.); [email protected] (A.K.) * Correspondence: [email protected]; Tel.: +41-44-635-5175

Received: 14 June 2017; Accepted: 15 August 2017; Published: 26 August 2017

Abstract: The Karakoram mountain range is well known for its numerous surge-type glaciers of which several have recently surged or are still doing so. Analysis of multi-temporal satellite images and digital elevation models have revealed impressive details about the related changes (e.g., in glacier length, surface elevation and flow velocities) and considerably expanded the database of known surge-type glaciers. One glacier that has so far only been reported as impacted by surging tributaries, rather than surging itself, is the 50 km long main trunk of Hispar Glacier in the catchment. We here present the evolution of flow velocities and surface features from its 2015/16 surge as revealed from a dense time series of SAR and optical images along with an analysis of historic satellite images. We observed maximum flow velocities of up to 14 m d−1 (5 km a−1) in spring 2015, sudden drops in summer velocities, a second increase in winter 2015/16 and a total advance of the surge front of about 6 km. During a few months the surge front velocity was much higher (about 90 m d−1) than the maximum flow velocity. We assume that one of its northern tributary glaciers, Yutmaru, initiated the surge at the end of summer 2014 and that the variability in flow velocities was driven by changes in the basal hydrologic regime (Alaska-type surge). We further provide evidence that Hispar Glacier has surged before (around 1960) over a distance of about 10 km so that it can also be regarded as a surge-type glacier.

Keywords: Karakoram; Hispar Glacier; surge; hydrology controlled; velocity; time series; Landsat; Sentinel-1; Corona

1. Introduction The Karakoram region is well known for its surge-type glaciers [1]. In particular in the central Karakoram recently or currently actively surging glaciers are abundant (e.g., [2–6]). During the active phase of a typical glacier surge, large amounts of ice are rapidly (by a factor of ten or more compared to normal flow velocities) transported from a higher so-called reservoir area to a lower receiving area. After the surge, in its quiescent phase, ice flow slows or becomes stagnant allowing ice to melt away over years to decades in the new ablation area while new snow and ice accumulate in the reservoir area, building up mass for a possible next surge, (e.g., [7]). In cases where a surging glacier interacts with another glacier (i.e., as either an ephemeral or permanent tributary), surface morphological features, such as distorted or tear-dropped moraines, sometimes develop. These have been used widely to infer past surge-activity and identify surge-type glaciers in satellite imagery, (e.g., [8–10]). The surges in the Karakoram show a wide range and mixture of characteristics regarding frontal advances (or not) and advance rates, surge duration, velocity increase, and mass change pattern (among others) that have recently been studied in detail using optical and SAR (Synthetic Aperture Radar) remote sensing data, (e.g., [3,5,11–14]). Glacier surges in the Karakoram also vary in regard to their interactions. For example, they can be related to isolated speed-up events within a single glacier

Remote Sens. 2017, 9, 888; doi:10.3390/rs9090888 www.mdpi.com/journal/remotesensing Remote Sens. 2017, 9, 888 2 of 18 that are difficult to distinguish from seasonal velocity variations, glaciers surging into each other as described above, mass waves travelling down-glacier faster than the surface flow velocity, the typical strong advances of the terminus over a short (months to years) or extended (decades) period of time, or a ‘response surge’ following flow-blocking release when the surge of a tributary glacier (blocking the flow of another glacier) comes to an end, (e.g., [4,6]). Some features used for identification of surge-type glaciers, such as looped or distorted moraines, can only be observed on larger glaciers with debris cover and thus might not always exist. This can make identification of surge-type glaciers in their quiescent phase difficult and biased, as the number of identified surge-type glaciers might increase with time and increasing information availability, not least through satellite data. Analysis of satellite image time-series (or aerial photography if available) and historic reports can be helpful to reveal possible former surges [1,4,8,15], but often it is required to use several criteria for a clear identification, (e.g., [5,16]). For example, some glaciers might only show a strong increase of flow velocity without corresponding terminus advance, (e.g., [17]), or others advance at rates similar to non-surge-type glaciers (less than 100 m y−1) but do so for more than 10 or 20 years (e.g., First Feriole in the Panmah region). As it is not yet clearly defined how a surge-type glacier can be identified (cf. the list of criteria summarized by [16]), their number for a specific region will also vary with the interpretation by the analyst. Hispar Glacier (HG) is one of the largest glaciers in the Karakoram that has reportedly so far not surged itself, (e.g., [1]), but is accounted for being impacted by surges of its tributaries. In this study we describe its 2015/16 surge using a time series of Landsat images (sensors MSS, TM, ETM+, and OLI) to derive morphological changes and pre-surge variability of flow velocity. In addition, SAR images from Sentinel-1 and RADARSAT-2 are exploited together with the optical data to derive a dense time series (about bi-weekly to monthly means) of flow velocities during the surge. Based on these results we hypothesize about a possible surge mechanism. By analyzing historic reconnaissance imagery from Corona, we also provide evidence that the middle part of HG experienced a surge around 1960.

2. Study Region Hispar Glacier is located in the central Karakoram and drains into the Hunza catchment in (Figure1, inset). Its main trunk is a nearly 50 km long (area about 500 km 2) linear stream of ice, flowing towards WNW from Hispar pass at 5150 m a.s.l. down to 3100 m. It is the dominant glacier in the near-linear main valley but other large surge-type glaciers (Trivor, Bualtar) drain further to the west into the main valley. A high mountain range (up to 7000 m) is protecting the ESE-WNW orientated valley in the south and the Hispar Muztagh (up to 7800 m) in the north. At Hispar pass, HG is connected to the third longest glacier in the world, the 65 km long Biafo Glacier, flowing to the East. Hispar glacier has a comparably tiny accumulation area that is not much wider than the tongue in the ablation region and is composed of three small basins. However, from the south about a dozen small tributaries nourish the tongue and from the north, three large tributary glaciers feed the main glacier trunk (Kunyang, Yutmaru and Khani Basa). The tributaries collectively contribute about a half of the total width of the main tongue and the looped/distorted medial moraines on the main glacier indicate former surge or fast-flow events. The three northern main tributaries of HG are complex glacier systems with multiple tributaries nourishing their respective main tributaries, i.e., their shape has a sort of fractal self-similarity. Due to the high relief of their surrounding ice-free rock walls, the ablation areas of the three tributaries are also debris-covered (with decreasing amounts from Kunyang in the west to Khani Bassa in the east). The lower part of the main tongue of HG is completely debris covered and characterized by numerous supra-glacial melt ponds as well as circular collapse structures, indicating the widespread down-wasting of the tongue. The latter has been confirmed in the study by [14] for the period 2000–2008 from differencing of digital elevation models (DEMs). Limited information is available about the climatic regime of the glacier studied as meteorological stations are sparse, their time series are often short (a few decades), and they are located at comparably Remote Sens. 2017, 9, 888 3 of 18 Remote Sens. 2017, 9, 888 3 of 18 low elevationsLimited [18 information]. Mean annual is available air temperature about the at theclimatic station regime of (1460 the mglacier a.s.l.) aboutstudied 80 as km to the west-southwestmeteorological stations of HG areare 15.8sparse,◦C[ their19], givestime series a value are of often about short−3 (a◦C few (lapse decades), rate − and0.007 they◦C are m− 1) at the midpointlocated at elevation comparably (4125 low m)elevations of HG. [18]. According Mean annual to [20 air] the temperature region receives at the station about Gilgit the same (1460 share m of precipitationa.s.l.) about in summer80 km to andthe west-southwest autumn as in winter of HG andare 15.8 spring, °C [19], i.e., gives it is influenceda value of about by both −3 °C the (lapse monsoon −1 and westerly’s.rate −0.007 °C Annual m ) at amountsthe midpoint are veryelevation low (4125 (e.g., m) in of Gilgit HG. 137According mm according to [20] the to region [19]) butreceives increase about the same share of precipitation in summer and autumn as in winter and spring, i.e., it is with elevation to about 1 m at 4500 m a.s.l [21]. A study by [22] indicated amounts in the order of influenced−1 by both the monsoon and westerly’s. Annual amounts are very low (e.g., in Gilgit 137 1.2 mmm a accordingin the accumulation to [19]) but increase region with of the elevation neighbouring to about Biafo1 m at Glacier 4500 m anda.s.l [[21].23] reportedA study by between [22] 1 and 2indicated m at Hispar amounts pass in (5060 the order m a.s.l.). of 1.2 m a−1 in the accumulation region of the neighbouring Biafo TakenGlacier together,and [23] reported this suggests between a 1 coldand 2 continental m at Hispar climatepass (5060 and m a.s.l.). that HG could, at least in parts, be poly-thermal.Taken together, A slight this increase suggests in a wintercold continental precipitation climate and and decrease that HG incould, summer at least temperature in parts, be has beenpoly-thermal. reported by A [18 slight] and increase confirmed in winter by [19 prec] usingipitation nearby and meteorologicaldecrease in summer stations. temperature This indicates has increasinglybeen reported favourable by [18] conditions and confirmed (in mass by [19] balance using terms)nearby for meteorological glacier survival. stations. The This related indicates balanced to slightlyincreasingly positive favourable mass budgets conditions over (in the mass 2000 balanc to 2010e terms) decade for haveglacier been survival. confirmed The related by several balanced studies that analysedto slightly glacier positive elevation mass budgets changes, over independent the 2000 to of2010 surge-type decade have glaciers been being confirmed in the by sample several or not, studies that analysed glacier elevation changes, independent of surge-type glaciers being in the (e.g., [14,24]). A comparison with historic maps and early photographs of the terminus reveal limited sample or not, (e.g., [14,24]). A comparison with historic maps and early photographs of the changesterminus in terminus reveal limited position changes of HG overin terminus the past positi 120on years. of HG This over also the indicates past 120 nearly years. balanced This also mass budgetsindicates over thisnearly time balanced period, mass and budgets several over decades this time before. period, and several decades before.

Figure 1. The location of the Hispar Glacier in the Karakoram is indicated on the inset (screen shot Figure 1. The location of the Hispar Glacier in the Karakoram is indicated on the inset (screen shot from Google Earth) with a black square. The annotated main image depicts the outlines of Hispar from Google Earth) with a black square. The annotated main image depicts the outlines of Hispar Glacier and its tributaries in white and of other glaciers in black. The dotted yellow lines denote Glaciercentre and lines its tributaries used for extraction in white of and flow of velocities other glaciers (here in in 200 black. m equidistance The dotted). yellow Glaciers lines mentioned denote in centre linesthe used text for are extraction named. The of main flow velocitiesimage is a false (here colour in 200 composite m equidistance). (bands 6, 5, Glaciers 4 as RGB) mentioned from Landsat in the 8 text are named.OLI acquired The mainon 9 October image is2013. a false colour composite (bands 6, 5, 4 as RGB) from Landsat 8 OLI acquired on 9 October 2013. 3. Data Sets and Image Processing Methods

Remote Sens. 2017, 9, 888 4 of 18

3. Data Sets and Image Processing Methods

3.1. Data Sets Two types of satellite images have been used in this study: (1) Optical images from the Landsat series and Corona to (a) illustrate and interpret historic as well as recent changes of HG in a more qualitative and geomorphologic sense, and (b) to derive flow velocities (only Landsat); and (2) time-series from SAR sensors (Sentinel-1/RADARSAT-2) that have been acquired during the surge were used to derive surface flow velocities with high temporal resolution. The satellite images used are listed in AppendixB Table A1 (animation), Table A2 (pre-surge) and Table A3 (main-surge). High-resolution (about 5 m) Corona scenes from 1965 and 1969 were used to identify the surface morphology of HG prior to 1977, when a Landsat MSS scene with very good quality is available. After 1990 Landsat scenes are available every 2–3 years. For the 2015/16 surge period a Landsat 7/8 time series could be obtained to follow the evolution of the surge visually (see Table A1). We have created an image series from these scenes (Supplemental File S1) that can be animated to reveal the dynamics of the surge [4]. As some images have fresh snow-cover, stripes from Landsat 7 SLC-off scenes, and clouds, all creating strong changes in contrast, it is recommended to focus viewing on the surging parts of the glacier tongue. All images are derived from the standardized image quicklooks provided by USGS (e.g., glovis.usgs.gov). They have a somewhat reduced quality (jpg-compression) compared to the original data but the advantage of homogenous contrast and colours, which facilitates visual interpretation of the animations. To derive flow velocities, the original panchromatic bands from Landsat 7 ETM+ and Landsat 8 OLI (15 m resolution) have been used. Landsat data were freely downloaded from the website earthexplorer.usgs.gov. The Sentinel 1 images were provided by ESA and obtained from the Sentinel Science Hub (scihub.copernicus.eu). The RADARSAT-2 scenes were provided by the Canadian Space Agency (see Acknowledgments). For comparison with current glacier extents and surface morphology, we have also analysed the historic map from the Bullock-Workmann expedition in 1908 that is available online. We used glacier outlines for the region as available from the Randolph Glacier Inventory (RGI) version 5.0 [25] to spatially constrain the analysis of the flow velocities, assign regions of stable ground for uncertainty assessment of flow velocities, and for visualisation of glacier extent.

3.2. Methods Feature tracking of 19 Landsat 7 ETM+ scenes and 14 Landsat 8 OLI scenes (panchromatic band with 15 m spatial resolution) acquired between July 2013 and January 2016 was performed using the CIAS software [26,27]. We used the orientation correlation function to extract displacements with a reference block size of 15 pixels (225 m), a search area size of 50 pixels (750 m) at a grid of 100 m. Table1 provides the parameters used for the tracking algorithm. Three RADARSAT-2 Wide Fine (RS-2 WF) scenes were acquired between December 2014 and February 2015. These repeat data were used to retrieve two glacier surface velocity maps using the offset tracking method of the GAMMA software [28]. The size of the correlation matching window was adjusted according to the image resolution and expected maximum displacements during the repeat pass cycle to 38 × 77 pixels in range and azimuth (see Table1). The resulting RS-2 velocity maps were then geocoded using the SRTM DEM to a 100 m resolution raster. A series of Sentinel-1 Interferometric Wide Swath (IWS) images were acquired between March 2015 and December 2016 along both ascending and descending orbits (AppendixB Table A3). The repeat Sentinel-1 IWS data were used to retrieve a time-series of surface velocity maps using the offset tracking method of the GAMMA software [28] (see Table1 for parameters). Slant-range and azimuth offset-fields were combined to retrieve horizontal surface velocity maps. These were geocoded using the SRTM DEM at a 100 m resolution raster. Remote Sens. 2017, 9, 888 5 of 18

Table 1. Specifications and processing parameters of Landsat 7 ETM+, Landsat 8 OLI, RADARSAT-2 and Sentinel-1 data. The spatial resolution of the velocity map is 100 × 100 m for all sensors.

Landsat ETM+ Landsat OLI RADARSAT-2 Sentinel-1 Mode Panchromatic Panchromatic Wide Fine IWS Scene coverage (km) 170 × 183 170 × 183 150 × 150 250 Pixel resolution (m) 15 × 15 15 × 15 10.6 × 5.2 3 × 22 m Matching window (pixel) 15 × 15 15 × 15 38 × 77 256 × 128 Search area (pixel) 50 × 50 50 × 50 38 × 77 256 × 128 Step size 100 m 100 m 9 × 19 pixel 40 × 10

The Corona images were only analysed qualitatively and not geocoded. All Landsat satellite images were used as is, i.e., terrain corrected by USGS (Level 1T). Mismatches and outliers were identified and removed from the velocity maps using different tests. For the Sentinel 1 and RADARSAT-2 data the signal-to-noise ratio had to be above 2.0 to result in a valid match. Landsat velocity maps were filtered with a standard deviation filter. Values are defined as outlier when the standard deviation of a pixel in a 3 × 3 window surrounding it exceeded 32 m. Moreover, in all maps values above 15 m d−1 (the maximum flow velocity) were removed. The accuracy of the derived velocity fields for Sentinel-1 was estimated from the displacement over stable terrain (using the RGI glacier outlines as a mask and restricting calculations to flat terrain. To follow the spatio-temporal development of flow velocities in more detail, we manually digitized three centre lines: one along the main glacier and two for its middle and eastern tributaries. These were converted to points with discrete IDs and a distance of 50 m (see Figure1). Their coordinates and IDs were exported to a csv file and used in a small script for automated extraction of values from the computed velocity fields. Resulting values were organized and further processed for visualisation. Values for the eastern tributary (Khani Bassa) are not shown as the analysis revealed that it did not participate in the surge studied here.

4. Results

4.1. Historic Development of Hispar Glacier from Landsat Data The surface changes of HG from the available historic satellite data from Landsat (sensors MSS/TM/ETM+/OLI) reveal some interesting details about its past development, all of which are not obvious from its stagnant terminus, geodetic volume changes or analysis of flow velocities. Figure2 shows a time series of the glacier surface from 1977 (MSS), 1990 (TM), 2013 (OLI) and 2015 (ETM+) with some characteristic features marked by arrows with different colours. Through the full near 40-year time series the terminus position has not changed (green arrow at top left). Characteristic surface features (distorted moraines) indicated by yellow, white and orange arrows move slowly down-glacier from 1977 to 2013, with a somewhat higher velocity in the upper glacier regions (larger displacement of the yellow arrow). In total, five of such marks are visible in the 1977 MSS image (Figure2a) indicating previous disturbances of the flow. A prominent moraine distortion (without arrow in the figure) at the confluence with Pumarikish Glacier (Figure2a) was destroyed by its 1989 surge [ 29] that is marked by a white circle on Figure2b. Another surge mark to the east of the confluences with Kunyang Glacier is still visible in the 1990 image but no longer present in 2013 (Figure2c). Further extracts from the time-series of Landsat images reveal that Kunyang Glacier started to surge in 2006 and advanced into HG in 2008 with some further advance until 2009. This surge is also visible in Figure9 of the study by [ 14] in which they calculated DEM differences between 1999 (from SRTM) and 2008 (from SPOT). In the 2013 image (Figure2c) the glacier is again in its quiescent phase, but the marks of the surge are still recognizable by its lobate form, covering about 3/4 of the width of HG. The good contrast in the 2013 image reveals the exposed northern lateral moraine (shorter black arrows), indicating that the glacier surface had been previously higher. Remote Sens. 2017, 9, 888 6 of 18

Remote Sens. 2017, 9, 888 6 of 18

Figure 2. Time series of the surface evolution of the lower part of Hispar Glacier as seen with the Figure 2. LandsatTime sensors series (a) of MSS the (1977), surface (b evolution) TM (1990), of ( thec) OLI lower (2013), part and of ( Hispard) ETM+ Glacier (2015). asArrows seen withand the Landsatcircles sensors point (toa) specific MSS (1977), surface (b )features TM (1990), discussed (c) OLI in (2013),the text. and Image (d) ( ETM+a) is a (2015).false colour Arrows composite and circles pointwith to specific MSS bands surface 6, 5, features4 as RGB, discussed images (b– ind) theuse text.the SWIR, Image NIR (a )and is a red false channel colour as compositeRGB. with MSS bands 6, 5, 4 as RGB, images (b–d) use the SWIR, NIR and red channel as RGB. 4.2. The 2015/16 Surge from Optical Images 4.2. The 2015/16When contrasting Surge from Opticalthe appearance Images of the autumn 2015 glacier surface (Figure 2d) with the Whenappearance contrasting in 2013 (Figure the appearance 2c), strong changes of the autumncan be recognized. 2015 glacier Most surface prominent (Figure is the2 d)reduced with the debris load due to the increased crevassing (indicated by the more “bluish” surface in regions appearance in 2013 (Figure2c), strong changes can be recognized. Most prominent is the reduced marked by the light blue arrows) for both HG and its tributary Yutmaru Glacier. The synchronous debrissurge load of due both to glaciers the increased is supported crevassing by the (indicated stable position by the ofmore the medial “bluish” moraine surface between in regions them marked(if by theYutmaru light blue had arrows)surged onto for bothHG, the HG moraine and its should tributary have Yutmaru been translated Glacier. and The deformed). synchronous One surgecan of both glaciers is supported by the stable position of the medial moraine between them (if Yutmaru had surged onto HG, the moraine should have been translated and deformed). One can also recognize the lateral expansion of the tongue and the increased ice thickness (the lateral moraine in the 2013 Remote Sens. 2017, 9, 888 7 of 18 Remote Sens. 2017, 9, 888 7 of 18 also recognize the lateral expansion of the tongue and the increased ice thickness (the lateral moraine imagein the disappeared),2013 image disappeared), along with the along forward with movement the forward of the movement surface marks of the (large surface orange marks and white(large arrows).orange and Further white downstream arrows). Further the surge downstream front can bethe identified surge front (light can bluebe identified line with (light half elliptical blue line shape with inhalf Figure elliptical2d). The shape lakes in west Figure of the 2d). orange The lakes arrow west in the of 2013 the imageorange have arrow disappeared. in the 2013 The image animation have ofdisappeared. the 2015/16 The Landsat animation images of (see the Table2015/16 A1 )Landsa clearlyt revealsimages a (see mass Table wave A1) with clearly a related reveals surge a frontmass travellingwave with down a related glacier surge in 2015/16.front travelling The surge down front glacier stopped in 2015/16. before The the surge confluence front withstopped Kunyang before Glacierthe confluence that is still with occupying Kunyang about Glacier 3/4 that of theis st widthill occupying of HG. about 3/4 of the width of HG.

4.3.4.3. FlowFlow VelocitiesVelocities before,before, duringduring andand afterafter thethe SurgeSurge InIn FiguresFigures3 –3–66 we we show show the the temporal temporal evolution evolution of flow of flow velocities velocities before before the surge the surge as derived as derived from time-seriesfrom time-series of Landsat of Landsat images images (Figures (Figures3 and4) and3 and during/after 4) and during/after the surge the from surge time from series time of Landsat series of andLandsat Sentinel-1 and Sentinel-1 images (Figures images5 (Figures and6). Results5 and 6). from Results RADARSAT-2 from RADARSAT-2 are not shown are not as theshown Landsat as the timeLandsat series time had series a better had quality. a better The quality. figures The show figur eitheres theshow full either flow fieldthe full from flow selected field imagefrom selected pairs of theimage time pairs series of (Figures the time3 andseries5) or(Figures mean values3 and along5) or mean a central values flow along line of a HGcentral in time-distance flow line of plotsHG in withtime-distance velocities plots colour with coded velocities (Figures colour4 and 6coded). In the (Figures latter two,4 and the 6). red-brown In the latter horizontal two, the linered-brown notes thehorizontal position line where notes the the Yutmaru position tributary where the joins Yutmaru the flow tributary of HG. joins the flow of HG.

FigureFigure 3.3. Pre-surgePre-surge variabilityvariability ofof flowflow velocitiesvelocities inin thethe summersummer ofof 20132013 ((aa)) andand ((bb)) andand 20142014 ((cc)) andand ((dd)) derivedderived from from Landsat Landsat 8 OLI.8 OLI. (e) Initiation(e) Initiation of the of surge the surge in late in autumn late autumn 2014 by 2014 the Yutmaru by the tributaryYutmaru (red)tributary and ((red)f) start and of the(f) fullstart surge of the in full winter surge 2014/15. in winter (g) is 2014/15. the same (g as) is (f )the but same with aas different (f) but colourwith a tabledifferent to better colour reveal table the to furtherbetter reveal acceleration the further of the acceleration flow shown of in the (h). flow Coordinates shown in are (h in). UTM43N.Coordinates are in UTM43N.

Remote Sens. 2017, 9, 888 8 of 18

The first two image pairs in Figure3 show the summer speed-up and slow-down in July and AugustRemote of 2013 Sens. (Figure 2017, 9, 8883a,b) and 2014 (Figure3c,d). Although difficult to see due to data voids,8 of 18 these velocity increases reach beyond the eastern tributary well into the accumulation area of HG. Notably, The first two image pairs in Figure 3 show the summer speed-up and slow-down in July and −1 the summerAugust speed-up of 2013 (Figure in 2014 3a,b) reached and 2014 peak (Figure velocities 3c,d). Although twice as difficult high as to insee 2013 due to (nearly data voids, 2 m these d instead −1 of 1 m dvelocity). Interestingly, increases reach Figure beyond3e shows the eastern an increase tributary in well flow into velocities the accumulation for the Yutmaru area of tributaryHG. in SeptemberNotably, 2014 the instead summer of speed-up a further in slowdown 2014 reached as peak in 2013.velocities This twice increase as high has as in seemingly 2013 (nearly activated 2 m the ice of HG,d−1 instead which of is 1 still m d− in1). itsInterestingly, summer-mode Figure 3e of shows increased an increase speed, in asflow Figure velocities3f shows for the flowYutmaru velocities exceedingtributary 2 m d in− 1Septemberfor both HG2014 andinstead Yutmaru of a further Glacier slowdown in November/December as in 2013. This increase 2014. has seemingly activated the ice of HG, which is still in its summer-mode of increased speed, as Figure 3f shows This result is shown again in Figure3g but with an extended colour scale (max. is now 4 m d−1) flow velocities exceeding 2 m d−1 for both HG and Yutmaru Glacier in November/December 2014. −1 to reveal theThis further result is flow shown acceleration again in Figure during 3g but Januarywith an extended 2015 to colour 4 m scale d (max.(Figure is now3h). 4 m The d−1) related time-distanceto reveal plot the in further Figure flow4 confirms acceleration the upduring and downJanuary of 2015 velocities to 4 m withd−1 (Figure additional 3h). The time related slices. From Octobertime-distance 2013 to January plot in 2014 Figure flow 4 confirms velocities the are up indeedand down below of velocities 0.4 m d with−1 (or additional 150 m a time−1). slices. The massive increaseFrom in flow October velocities 2013 to after January the 2014 surge flow started velocities in October are indeed 2014 below is well 0.4 m visible d−1 (or for 150 the m a last−1). The two image massive increase in flow velocities after the surge started in October 2014 is well visible for the last pairs from December 2014 and January 2015. From these plots it seems that the surge started beyond two image pairs from December 2014 and January 2015. From these plots it seems that the surge the confluencestarted beyond region withthe confluence Yutmaru region (at km with 31) andYutm increasedaru (at km flow 31) velocitiesand increased occurred flow velocities also backwards towardsoccurred the accumulation also backwards region. towards the accumulation region.

Figure 4.FigureEvolution 4. Evolution of flow of flow velocities velocities in in a a time-distance time-distance plot plot along along the theflow flow line of line HG. of The HG. black The black dotted line across all dates shows the position of the confluence with the Yutmaru tributary. dotted line across all dates shows the position of the confluence with the Yutmaru tributary. In Figure 5 the evolution of surface flow velocities from March 2015 to June 2016 is shown based In Figureon time5 series the evolution of Sentinel-1 of images. surface Velocities flow velocities increased from to about March 6 m 2015 d−1 in to April June 2015 2016 (Figure is shown 5b) based −1 on timeand series 7 m of d Sentinel-1 until mid-May images. (Figure Velocities 5c). Flow increased velocities toof aboutthe Yutmaru 6 m d −tributary1 in April are 2015already (Figure much5 b) and lower then and it looks like the surge is just originating in the centre of HG. Flow velocities 7 m d−1 until mid-May (Figure5c). Flow velocities of the Yutmaru tributary are already much lower downstream of the confluence area nearly doubled to about 14 m d−1 (about 5 km y−1) until the end of then andMay it looks(Figure like 5d) theand surgedropped is justto 5 originatingm d−1 (Figure in5e) the and centre less than of HG.2 m d Flow−1 (Figure velocities 5f) by the downstream end of of −1 −1 the confluenceJuly and area August, nearly respectively. doubled toDuring about winter 14 m and d spring(about 2016 5 km flow y velocities) until the for endthe main of May glacier (Figure 5d) and droppedincreased to 5again m d to− 1about(Figure 7 m5 de)−1 and(Figure less 5g,h) than before 2 m dthey−1 sharply(Figure dropped5f) by the to less end than of July 1 m andd−1 by August, respectively.July 2016. During This dual winter surge and with spring the sharp 2016 drop flow can velocitiesalso be seen for in the the time-distance main glacier plot increased of Figure 6. again to −1 about 7In m this d− 1figure,(Figure we 5haveg,h) restricted before they maximum sharply velocities dropped to 10 to mless d to than reveal 1 mdetails d −1 inby the July second 2016. surge This dual pulse. Apart from the sudden drop in flow velocities the most interesting aspect is a very subtle surge withone—the the sharp advance drop of the can surge also front. be seen For the in thethree time-distance image pairs acquired plot of at Figure the beginning6. In this of figure,May, end we have −1 restrictedof maximumMay and end velocities of June of to 2015 10 mthe d zoneto with reveal speeds details of around in the 6 secondm d−1 (yellow surge area pulse. in Figure Apart 6) from the suddenmoved drop inin about flow velocities25 days by about the most 2–2.5 interestingkm or about aspect90 m d−1 is. The a very rapid subtle advance one—the of the surge advance front of the surge front. For the three image pairs acquired at the beginning of May, end of May and end of June of 2015 the zone with speeds of around 6 m d−1 (yellow area in Figure6) moved in about 25 days by about 2–2.5 km or about 90 m d−1. The rapid advance of the surge front could thus be related to a kinematic wave or some other mechanism near the glacier bed that has activated the flow. Remote Sens. 2017, 9, 888 9 of 18

Remotecould Sens.thus2017 be, 9related, 888 to a kinematic wave or some other mechanism near the glacier bed that9 ofhas 18 activated the flow.

Figure 5. (a–h) Temporal evolution of flow velocities from spring 2015 to summer 2016 derived from Figure 5. (a–h) Temporal evolution of flow velocities from spring 2015 to summer 2016 derived from Sentinel-1. Highest velocities are reached by the end of May 2015 (d). After a substantial velocity Sentinel-1. Highest velocities are reached by the end of May 2015 (d). After a substantial velocity decrease in autumn 2015 (f), velocities increased again in 2016 (g) and (h). Please note the different decrease in autumn 2015 (f), velocities increased again in 2016 (g) and (h). Please note the different colour scale in Figure5 (max. 15 m d −1) and Figure6 (max. 10 m d −1). colour scale in Figure 5 (max. 15 m d−1) and Figure 6 (max. 10 m d−1).

The scatterplotscatterplot showingshowing flow flow velocities velocities against against distance distance from from the the terminus terminus and and colour-coded colour-coded for timefor time (Figure (Figure7) confirms 7) confirms this evolution this evolution but shows but shows the rapid the advancerapid advance of the surgeof the front surge more front clearly. more Theclearly. point The separating point separating the regions the regions of fast and of fast slow an flowd slow rests flow more rests or more less ator the less same at the place same (at place km 15(at) − − untilkm 15) April until 2015. April Flow 2015. velocities Flow velocities increased increased to about to 2.5 about km a2.51 km(about a−1 (about 7 m d 71 )m behind d−1) behind this point this beforepoint before it advanced it advanced by nearly by 5nearly km in 5 2 km months. in 2 months. The scatter The plot scatter also plot shows also the shows drop inthe flow drop velocities in flow wherevelocities the where two tributaries the two tributaries Yutmaru (kmYutmaru 28) and (km Khani 28) and Basa Khani (km Basa 35) join (km the 35) flow. join Thethe flow. part betweenThe part thebetween two confluence the two confluence points (upstream points (upstream of Yutmaru) of Yutm takesaru) part takes in thepart surge in the (though surge (though on a lower on a lower level) whereaslevel) whereas the part the upstream part upstream of Khani of Basa Khani confluence Basa confluence did not. During did not. the During period withthe period the highest with flow the − − velocitieshighest flow there velocities is a ±500 there m a is1 a(1.4 ±500 m dm 1a)−1 variability (1.4 m d−1) of variability flow velocities of flow between velocities km 15between and 28, km even 15 inand this 28, smoothed even in this display smoothed (averaged display over (averaged 200 m bins). over However, 200 m bins in). view However, of potential in view measurement of potential uncertaintiesmeasurement this uncertainties variability this is likely variability within is the likely error within bounds the and error should bounds thus and not beshould over interpreted.thus not be Inover contrast interpreted. to the dynamicsIn contrast of to the the upper dynamics glacier of parts the thatupper are glacier influenced parts by that the are surge, influenced flow velocities by the − forsurge, the flow nearly velocities stagnant for lower the nearly part of stagnant the tongue lower are part below of 30the m tongue a 1. are below 30 m a−1.

Remote Sens. 2017, 9, 888 10 of 18 Remote Sens. 2017, 9, 888 10 of 18 Remote Sens. 2017, 9, 888 10 of 18

FigureFigure 6. Evolution 6. Evolution of flow of flow velocities velocities during duringthe the surgesurge in in a a time-distance time-distance plot plot along along the flow the flowline of line of FigureHG (cf. 6. Figure Evolution 4). The of flowthick velocities black-dotted during line the across surge all in dates a time-distance shows the positionplot along of thethe flow confluence line of HG (cf. Figure4). The thick black-dotted line across all dates shows the position of the confluence with HGwith (cf. the Figure Yutmaru 4). tributary.The thick Theblack-dotted two separate lined across fast flow all periodsdates shows are clearly the position visible. of the confluence the Yutmaruwith the tributary. Yutmaru tributary. The two The separated two separate fast flowd fast periodsflow periods are clearlyare clearly visible. visible.

Figure 7. Annotated time-series of flow velocities vs. distance from the terminus along the central FigureFigureflow 7. Annotated line 7. ofAnnotated Hispar time-series Glacier, time-series colour of of flow codedflow velocities velocities for date andvs. distance distanceusing a 200from from m the sampling. the terminus terminus The along location along the central of the the central flow lineflowtwo of tributaries line Hispar of Hispar Glacier, is indicated. Glacier, colour Thecolour coded vertical coded for double for date da arro andte andw using is using indicating a a 200 200 m them sampling. sampling. increase in The velocity location while of of the the two tributariestwosurge tributaries is front indicated. was is stable. indicated. The Note: vertical The 3650 vertical double m y−1 equalsdouble arrow a arro flow is indicatingw velocity is indicating of the 10 the increasem d increa−1. se in in velocity velocity while while thethe surge surge front was stable. Note: 3650 m y−1 equals a flow velocity of 10 m d−1. front was stable. Note: 3650 m y−1 equals a flow velocity of 10 m d−1. Flow velocities along the profile of the Yutmaru tributary change in the same way as the main glacierFlow but velocities are in general along somewhat the profile lower of the as Yutmaru can be seen tributary in Figure change 3. To in get the a bettersame wayoverview as the on main the Flowglaciertemporal velocities but development are in along general theof somewhat flow profile velocities oflower the for as Yutmaru canthe mabe sineentributary glacier in Figure and change 3.the To two get in tributaries,a thebetter same overview we way have ason also thethe main glaciertemporalcalculated but are indevelopment velocity general averages somewhat of flow along velocities lower the flow asfor lines,can the bema excluding seenin glacier in the Figure and data the3 .gaps two To get andtributaries, a starting better we overviewat kmhave 14 also for on the temporalcalculatedthe flow development line velocity on HG. averages of When flow plotting along velocities the these flow for averaged lines, the main excluding velocities glacier the for anddata the the gapsthree two and major tributaries, starting flow linesat km we against 14 have for also calculatedthetime flow (Figure velocity line 8),on averages theHG. main When alongvariability plotting the theseemerges. flow averaged lines, Flow excluding velocities theforshow the data a threecontinuous gaps major and flowacceleration starting lines atagainst from km 14 for time (Figure 8), the main variability emerges. Flow velocities show a continuous acceleration from the flow line on HG. When plotting these averaged velocities for the three major flow lines against time (Figure8), the main variability emerges. Flow velocities show a continuous acceleration from Autumn 2014 to May 2015 for both HG and the Yutmaru tributary followed by a massive drop in Summer 2015. Velocities stayed at this level until January 2016 for HG but slightly increased for Remote Sens. 2017, 9, 888 11 of 18 Remote Sens. 2017, 9, 888 11 of 18

Autumn 2014 to May 2015 for both HG and the Yutmaru tributary followed by a massive drop in Yutmaru.Summer Mean 2015. velocities Velocities increased stayed at forthis HGlevel inuntil February January 20162016 for and HG stayed but slightly at this increased level before for they suddenlyYutmaru. dropped Mean in velocities June 2016. increased Yutmaru for hasHG in about February constant 2016 flowand stayed velocities at this from level Decemberbefore they 2015 to May 2016.suddenly Although dropped the in mean June 2016. values Yutmaru given has in about Figure co6nstant are somewhat flow velocities arbitrary, from December one can 2015 roughly to say that averagedMay 2016. flow Although velocities the mean of the values Yutmaru given in tributary Figure 6 are at somewhat and after arbitrary, the peak one in can May roughly 2015 say are about half ofthat the HGaveraged values. flow The velocities eastern of the tributary Yutmaru (Khani tributary Basa) at and shows after athe small peak increasein May 2015 in floware about velocity in half of the HG values. The eastern tributary (Khani Basa) shows a small increase in flow velocity in Dec 2014,Dec but 2014, constantly but constantly declined declined afterwards. afterwards. At At a a closer closer view, view, therethere is a little little bit bit of of up up and and down down that is synchronousthat is synchronous with the with variability the variability of the of two the other two other glaciers. glaciers.

FigureFigure 8. Averaged 8. Averaged flow flow velocities velocities for for the the mainmain glac glacierier and and its itstwo two tributaries tributaries (E: Khani (E:Khani Basa, W: Basa, W: Yutmaru) at different points in time. For the main glacier only velocities between km 10 and 28 (see Yutmaru) at different points in time. For the main glacier only velocities between km 10 and 28 Figure 7) were averaged. (see Figure7) were averaged. 4.4. Accuracy of the Derived Flow Velocities 4.4. Accuracy of the Derived Flow Velocities The displacement over stable terrain for Sentinel-1 gives on average values of about 13 m a−1 Thewith displacement a standard deviation over stable of 20 terrainm a−1 (18.3 for for Sentinel-1 descending, gives 21.9on for averageascending values orbits) ofin aboutthe mean 13 m a−1 with a standardover all processed deviation scenes. of 20 mThis a− 1is(18.3 about for two descending, orders of 21.9magnitude for ascending smaller than orbits) the in observed the mean over displacements and should thus have no impact on the results. Similarly, a recent study by [30] found all processed scenes. This is about two orders of magnitude smaller than the observed displacements in comparison to higher resolution sensors a displacement error of about 30 m a−1 for Sentinel-1 and shouldinterferometric thus have wide no impactswath (IWS) onthe images results. with Similarly,12 days repeat, a recent and 15 study m a−1 byfor [RADARSAT-230] found in comparisonWide − to higherUltra resolution Fine (WUF) sensors images aover displacement a 24 days period. error Uncertainties of about 30 for m aLandsat-derived1 for Sentinel-1 velocities interferometric are wide swathsimilar (IWS) to previous images investigations with 12 days (e.g., repeat, [31–33]), and revealing 15 m a− 1displacementfor RADARSAT-2 accuracies Wide of one Ultra pixel Fine or (WUF) imagesbetter, over i.e., a 24 15 days m for period.Landsat 8, Uncertainties or 1 m d−1 for 16-day for Landsat-derived repeat [27,34]. velocities are similar to previous investigations (e.g., [31–33]), revealing displacement accuracies of one pixel or better, i.e., 15 m for 5. Discussion Landsat 8, or 1 m d−1 for 16-day repeat [27,34]. 5.1. Interpretation of the Observations 5. Discussion Based on the available optical and SAR datasets, the observed geomorphometric changes, and 5.1. Interpretationthe spatio-temporal of the Observations variability of the flow velocities, it seems that the 2015/16 surge of HG started at the confluence region of HG and Yutmaru glacier in October 2014, with steadily increasing flow Basedvelocities on the (up available to 5 km a− optical1 or 14 m and d−1) until SAR May datasets, 2015. At the about observed the same geomorphometric time, the previously changes, stagnant and the spatio-temporalsurge front variability started to move of the forward flow velocities, by about 5 itkm seems in about that two the months. 2015/16 Flow surge velocities of HG quickly started at the confluencedropped region by about of HG one and half Yutmaru in summer glacier 2015 inand October stayed 2014,at this withlevel steadilyuntil winter increasing 2015 before flow they velocities slightly increased again in spring 2016. The final drop in speeds back to pre-surge values was strong (up to 5 km a−1 or 14 m d−1) until May 2015. At about the same time, the previously stagnant surge and fast (see Figure 8). front started to move forward by about 5 km in about two months. Flow velocities quickly dropped by about one half in summer 2015 and stayed at this level until winter 2015 before they slightly increased again in spring 2016. The final drop in speeds back to pre-surge values was strong and fast (see Figure8). Taken together, this seems to be a surge of Alaska-type that is controlled by the hydrologic regime, (e.g., [7,35]). The build-up during winter 2014/15 points to an acceleration due to increased basal water Remote Sens. 2017, 9, 888 12 of 18 pressure resulting from an ineffective basal drainage system (linked cavities). This pressure could have been strongly reduced after basal flow became more efficient during the summer (connected channels), causing the first strong decrease in flow velocity. However, velocities remained high during summer/autumn 2015 when basal sliding was maybe increased due to water generated by friction from the high flow velocities, additional meltwater input, or the water drainage system constantly being destroyed by the high basal deformation. The second winter/spring increase of flow velocities indicates that basal drainage was getting even more inefficient again and water pressure increased, before drainage changed again to the more efficient channelized mode in summer 2016, terminating the surge. A speed-up in the flow during winter and spring has also been observed at other surging glaciers, (e.g., [5,7,15,36]) and hints to a common hydrologic cause for the observed timing of the surge. The observed seasonal variability before the surge is much lower in magnitude, influenced larger parts of the main glacier and had a different timing (highest velocities in summer). Regarding the double peak of flow velocities and the sudden drop at the end of the surge (see Figure8) we find some resemblance with the 2015 surge of Kyagar Glacier which is located to the north east in the Shaksgam valley and described by [37]. However, for Kyagar Glacier maximum velocities were much lower (2 m d−1) and reached in September. So, its surge mechanism might be different. A speed up in flow during winter was also observed by [38] for the West Kunlun Glacier, but we assume that the surge mechanism is different here as the glacier is at least poly-thermal if not cold based. The surge of HG started in October 2014 during the autumn deceleration phase and was probably initiated by a flow acceleration (or maybe a surge?) of Yutmaru Glacier (Figure3e). In contrast to other glaciers where surging tributaries deform medial moraines of a main glacier (e.g., Panmah or Skamri glaciers), the acceleration of Yutmaru did not deform the medial moraine of HG. Instead, HG immediately started to surge itself with about the same flow velocity (Figure3f,h). We speculate that the reason for this quick response is related to its still higher than normal summer flow velocity with considerable lubrication at the bed. The region with the increased flow velocities is identical with the region of faster flow before the surge (cf. Figure3e,f). But already in January 2015 the region upstream of the confluence region had much lower flow velocities. This early ‘decoupling’ can also be seen in Figure8. For the Yutmaru tributary the decoupling took place by the end of April 2015 (Figure5c) before the maximum flow velocities were reached. Considering that Yutmaru Glacier only provided a small push to HG before its surge started suggests that HG was ready to surge before, but a trigger was missing. It is also noteworthy that for a considerable time of flow acceleration (1/2 year) there was no forward movement of the surge front. This happened only after maximum flow velocities were reached in May/June 2015. By then, a considerable amount of ice accumulated upstream before it could finally break through and the mass wave visible in the satellite animation started to travel downward at high velocities. There is maybe an obstacle at the bed at km 15 that is difficult to pass.

5.2. Is the Surge Unprecedented? Going further back in time, there are no reports of a surging HG from the 19th and 20th century [1] and early topographic maps of the glacier surveyed in 1892 by W. Martin Conway and in 1908 by the Bullock-Workmann expedition do also not show any signs of irregular flow. Medial moraines are lined up in parallel and overall the glacier seems to have less debris cover than nowadays. However, there is one source indicating a possible previous surge of the main HG, a Corona image from 1965 (Figure9). This image with a spatial resolution of a few metres shows a chaotic and highly crevassed surface typical of a glacier surface a few years after a surge. On that image one can also see that the Kunyang Glacier tributary—that is now occupying 3/4 of the valley width—was nearly completely pushed away and occupied less than 20% of the width of the main valley. A well visible surge front is indicating that this ‘surge’ reached about 4 km further downstream compared to the 2015/16 surge but has also not reached the terminus. About four years later, another Corona image reveals an impressive change of the surface morphology (AppendixA, Figure A1). Whereas the extent of the former surge is still well visible, the surface is Remote Sens. 2017, 9, 888 13 of 18

RemoteAbout Sens. 2017 four, 9, 888years later, another Corona image reveals an impressive change of the surface13 of 18 morphology (Appendix A, Figure A1). Whereas the extent of the former surge is still well visible, the surface is strongly smoothed and the previous crevasses are barely visible. Instead, the entire surface isstrongly covered smoothed with (partly and thewater-fill previoused) crevassesdepressions are and barely sinkholes, visible. Instead,indicating the a entiremassive surface down-wasting is covered withand (partlycollapse, water-filled) which is depressionsa typical post-surge and sinkholes, phenomenon indicating when a massive the down-wastingsurged ice mass and becomes collapse, whichstagnant. is a The typical limited post-surge movement phenomenon after the surge when is the also surged supported ice mass by becomes three surge stagnant. marksThe (loops limited and distortedmovement moraines) after the that surge are is in also 1977 supported (MSS image by threein Figure surge 2a) marks in the (loops same position and distorted as in 1969 moraines) (they arethat located are in 1977 in front (MSS of imagePumarikish in Figure and2a) Kunyang in the same glaciers). position Interestingly, as in 1969 (theyKunyang are located Glacier in in front 1977 of is againPumarikish occupying and Kunyang about 50% glaciers). of the Interestingly,valley width, Kunyangindicating Glacier that the in 1977down-wasting is again occupying post-surge about ice mass50% of of the Hispar valley Glacier width, could indicating be easily that thepushed down-wasting away. The post-surge Corona image ice mass from of 1969 Hispar (Figure Glacier A1) could also showsbe easily a lobate, pushed dual-coloured away. The Corona surge imagemark before from 1969 the su (Figurerge front, A1) indicating also shows that a lobate, this former dual-coloured surge (it mightsurge markhave beforetaken place the surge around front, 1960, indicating i.e., 55 thatyears this before former the surge current (it mightsurge) have might taken have place sheared around of 1960,the lower i.e., 55 part years of a before southern thecurrent tributary, surge) likely might from have Gandar sheared Glacier of the (see lower Figure part 2a), of a which southern both tributary, have a dual-shadedlikely from Gandar moraine. Glacier In the (see latter Figure case,2 thea), whichsurge would both have have a also dual-shaded started at moraine. km 15 and In must the latter have travelledcase, the surgeat least would 9 km have downstream. also started Finally, at km 15a recent and must study have by travelled [39] showed at least elevation 9 km downstream. differences fromFinally, 1973 a recent to 1999 study using by DEMs [39] showed derived elevation from Hexa differencesgon KH-9 from stereo 1973 pairs to 1999and SRTM using DEMsfor the derived Hunza catchment.from Hexagon In this KH-9 study stereo HG pairsshows and a strongly SRTM for down the-wasting Hunza catchment. middle section In this that study is compliant HG shows with a thestrongly extent down-wasting of the surge middle and the section typical that ispost-s complianturge down-wasting with the extent described of the surge before and thefrom typical the post-surgemorphological down-wasting change. So describedthe 2015/16 before surge from of HG the morphologicalseems not unprecedented change. So thebut2015/16 has at least surge one of precursor.HG seems not unprecedented but has at least one precursor.

Figure 9. Annotated Corona imageimage fromfrom 19651965showing showing details details of of the the previous previous surge. surge. For For comparison comparison a ascreenshot screenshot of of the the same same region region in in 2013 2013 from from Google Google Earth Earth is is shown shown in in the the inset. inset.

The teardrop-shaped moraine distortions visible in Figures 2a and A1 indicate that additional The teardrop-shaped moraine distortions visible in Figures2a and A1 indicate that additional surges of HG should have happened in the past as each loop has to be created and pushed forward surges of HG should have happened in the past as each loop has to be created and pushed forward by a separate surge (or fast flow event). The only obvious glacier upstream of Kunyang being able to by a separate surge (or fast flow event). The only obvious glacier upstream of Kunyang being able create the flow distortions of the main glacier is Yutmaru. The three surge marks suggest that to create the flow distortions of the main glacier is Yutmaru. The three surge marks suggest that Yutmaru had at least three periods with unstable flow that slightly deformed the medial moraine Yutmaru had at least three periods with unstable flow that slightly deformed the medial moraine and—different from the here-reported event—might have partly blocked the flow of HG. and—different from the here-reported event—might have partly blocked the flow of HG. Considering Considering their elongated/stretched shape and the very low flow velocities of HG during its their elongated/stretched shape and the very low flow velocities of HG during its quiescent phase quiescent phase (see Figure 2), it can be assumed that HG has responded to this partial (see Figure2), it can be assumed that HG has responded to this partial flow-blocking also with a surge flow-blocking also with a surge (or fast-flow event) that pushed the surge marks forward into their (or fast-flow event) that pushed the surge marks forward into their current elongated (tear-drop) shape. current elongated (tear-drop) shape. Apart from the 1960s event suggested by the Corona images we Apart from the 1960s event suggested by the Corona images we do not know when these former surges do not know when these former surges might have happened. For comparison, the Sentinel-2A might have happened. For comparison, the Sentinel-2A image in Figure A2 shows how HG looks image in Figure A2 shows how HG looks shortly after the current surge ended. shortly after the current surge ended.

Remote Sens. 2017, 9, 888 14 of 18

5.3. Interpretation of Observations Our interpretation of the observed temporal evolution of flow velocities and geomorphological evidence is largely based on a translation of what has been reported and observed in previous studies to the case of HG. We think that the assumption that changes in basal hydrology had a major impact on the variability in flow velocities is robust and not unexpected, as this has been reported for other surging glaciers with a sub-seasonal analysis of flow velocities [15,32]. Our assumption that HG was ready to surge and just “waited” for a trigger to start it is more speculative, but we think reasonable in view of the observations. However, our interpretation does not necessarily mean that the surge mechanism for HG works exactly as we speculate. Further research and additional observations (e.g., elevation changes) would be required to improve on the current (limited) understanding, in particular when considering that other surging glaciers in the region show a very different behaviour (e.g., the continuous advance over 15 years of First Feriole Glacier). Our speculation about former surges of HG and its Yutmaru tributary are based on analysing time series of historic satellite images (Corona, Landsat). They are supported by multiple lines of evidence and are fully compliant with what has been observed elsewhere in terms of surge marks, looped moraines, crevasse decay patterns and surface elevation changes. We thus think our interpretation is robust in this regard. Together with Skamri Glacier to the north-east, HG is thus one of the largest surging glacier in the Karakoram. It is likely that further surge-type glaciers will be identified in the future based on their sudden change in surface flow, and new interpretation of surge marks and historic evidence. Thereby, the high temporal resolution of the current satellite coverage (Sentinel-1/2 every 5–6 days, Landsat 7/8 every 8 days) will likely also help reveal potential surge mechanisms and thus improve our understanding of this fascinating phenomenon.

6. Conclusions We presented a detailed description of the 2015/16 surge of Hispar Glacier as derived from the qualitative interpretation of optical image time series (Corona, Landsat) and the quantitative determination of flow velocities from both optical (Landsat 7 and 8) and SAR images (Sentinel-1 time series). Whereas the optical images allowed us to obtain pre-surge variability and the onset of the surge, the dense time series (partly every 2 weeks) of Sentinel-1 images allowed us to follow the spatio-temporal development of the surge in detail. The surge started in autumn 2014 and peaked with surface flow velocities of up-to 5 km a−1 (14 m d−1) over a short period in May/June 2015. During summer 2015 maximum velocities decreased substantially before they increased again in winter 2015/16. In June 2016 the surge ended abruptly. We assume that the variability of velocities over the full almost two-year duration of the surge are related to changes in the basal hydrologic regime and would thus classify the surge as Alaskan type. The most interesting aspects of the surge are: (1) the near-synchronous surge of HG and its tributary Yutmaru, (2) the stationary surge front (for 6 months), and (3) the high velocities of the surge front (up to 90 m d−1) once it started moving down-glacier. We interpret an end-of-summer fast-flow event of the Yutmaru tributary as a trigger of the main HG surge that was likely “ready to go” before. We do not know what kept the surge front stationary for half a year but interpret its rapid advance afterwards as a kinematic wave travelling down-glacier about six times faster than the maximum flow velocity. So far, surges have only been reported for some of the HG tributaries (Kunyang, Pumarikish) but not for HG itself. We have shown here that HG has very likely also surged around 1960 and maybe also before. None of these surges seem to have reached the terminus that has been stationary for >120 years according to the literature and historic maps. The high temporal resolution of satellite images now available will likely reveal similar details of surge evolution for other glaciers and contribute to an improved understanding of glacier surges.

Supplementary Materials: The following are available online at www.mdpi.com/2072-4292/9/9/888/s1, Animation S1: Hispar Glacier surge (zip file with individual images). Remote Sens. 2017, 9, 888 15 of 18

RemoteAcknowledgments: Sens. 2017, 9, 888 This study has been performed in the framework of the ESA project Glaciers_cci15 of 18 (4000109873/14/I-NB). T.S. was funded by the Norwegian Space Centre as part of European Space Agency’s PRODEX program (C4000106033), and the European Union FP7 ERC project ICEMASS (320816). A.K. also Acknowledgments:received funding fromThis the study European has been Union performed FP7 ERC in project the framework ICEMASS of (320816) the ESA and project the ESA Glaciers_cci project (4000109873/14/I-NB). T.S. was funded by the Norwegian Space Centre as part of European Space Agency’s Glaciers_cci (4000109873/14/I-NB). RADARSAT-2 data was provided by the Canadian Space Agency under the PRODEX program (C4000106033), and the European Union FP7 ERC project ICEMASS (320816). A.K. also proposalreceived fundingSOAR-EI from 5166. the Glacier European outlines Union were FP7 obtain ERC projected from ICEMASS the Randolph (320816) Glacier and Inventory the ESA project (RGI5.0). Glaciers_cci (4000109873/14/I-NB). RADARSAT-2 data was provided by the Canadian Space Agency under the proposal Author Contributions: F.P. conceived and designed the study, wrote the paper, and analysed the results. T.S. SOAR-EI 5166. Glacier outlines were obtained from the Randolph Glacier Inventory (RGI5.0). processed all Sentinel-1 data, T.S. processed most of the Landsat and all RADARSAT-2 data and created the Authorvelocity Contributions:plots in FiguresF.P. 3–6, conceived A.K. processed and designed further Landsat the study, data. wrote All author the paper,s contributed and analysed to the the interpretation results. T.S. processed all Sentinel-1 data, T.S. processed most of the Landsat and all RADARSAT-2 data and created the of the results and the writing and editing of the manuscript. velocity plots in Figures3–6, A.K. processed further Landsat data. All authors contributed to the interpretation of Conflictsthe results of and Interest: the writing The authors and editing declare of the no manuscript.conflict of interest. Conflicts of Interest: The authors declare no conflict of interest. Abbreviations Abbreviations The following abbreviations are used in this paper: The following abbreviations are used in this paper: a.s.l. above above sea sealevel level csv comma separated values csv comma separated values DEM Digital Elevation Model DEM Digital Elevation Model ETM+ Enhanced Thematic Mapper Plus ETM+ Enhanced Thematic Mapper Plus HG Hispar Glacier HGRADAR Hispar RAdio Glacier Detection And Ranging RADARRGB RAdio Red, Detection Green, and And Blue Ranging RGBMSS Red, Multispectral Green, and ScannerBlue MSSSAR Multispectral Synthetic Aperture Scanner Radar SARSLC Synthetic Scan-line-corrector Aperture Radar SLCSRTM Scan-line-corrector Shuttle Radar Topography Mission SRTMTM Shuttle Thematic Radar Mapper Topography Mission TMUSGS Thematic United Mapper States Geological Survey USGS United States Geological Survey Appendix A Appendix A

Figure A1. Corona image from 1969 showing the closure of crevasses, multiple surface lakes and Figure A1. Corona image from 1969 showing the closure of crevasses, multiple surface lakes and collapse structures typical of a down-wasting glacier surface. Three dark-banded surge marks from collapse structures typical of a down-wasting glacier surface. Three dark-banded surge marks from previous surges can be seen in the brighter part of the surface debris cover (white arrows). The black previous surges can be seen in the brighter part of the surface debris cover (white arrows). The black arrow is pointing to the dual-shaded surge mark that might have been sheared-off from Gandar arrow is pointing to the dual-shaded surge mark that might have been sheared-off from Gandar glacier glacier (dashed grey arrow). (dashed grey arrow).

Remote Sens. 2017, 9, 888 16 of 18 Remote Sens. 2017, 9, 888 16 of 18

FigureFigure A2. A2. TheThe middle middle part ofof Hispar Hispar Glacier Glacier at theat the end en ofd the of surge. the surge. The surge The front surge (arrow) front is(arrow) located is locatedbefore before the confluence the confluence region region with Kunyang with Kunyang Glacier. Glacier. Sentinel 2ASentinel image 2A from image 20 July from 2016 20 withJuly bands2016 with 8, bands4, and 8, 34, asand RGB 3 as (showing RGB (showing vegetation vegetati in red).on Copernicusin red). Copernicus Sentinel data Sentinel 2016. data 2016.

Appendix B

AppendixTable B A1. Scenes used for the time series in the supplement (ID 1 to 12) and Figure2 (IDs 1, 8, 13, 14). DoY is Day of Year. Table A1. Scenes used for the time series in the supplement (ID 1 to 12) and Figure 2 (IDs 1, 8, 13, 14). DoYID is Day Sensorof Year. Date DoY ID Sensor Date DoY 1ID L8Senso OLIr 9 OctoberDate 2013 282DoY ID 8 Senso L7 ETM+r 20 AugustDate 2015 DoY 232 2 L8 OLI 25 August 2014 237 9 L8 OLI 13 September 2015 256 1 L8 OLI 9 October 2013 282 8 L7 ETM+ 20 August 2015 232 3 L8 OLI 26 September 2014 269 10 L8 OLI 15 October 2015 288 42 L8 L8 OLI OLI 16 25 January August 2015 2014 16 237 11 9 L8 L8 OLI OLI 13 September 10 May 2016 2015 256 131 53 L8 L8 OLI OLI 26 21 September March 2015 2014 80 269 1210 L8 L8 OLI OLI 15 October 26 May 2015 2016 288 147 64 L8 L8 OLI OLI 1622 AprilJanuary 2015 2015 112 16 1311 L8 OLI MSS 10 2 May August 2016 1977 131 214 75 L8 L8 OLI OLI 21 9 June March 2015 2015 160 80 1412 L8 OLI TM 26 7 May August 2016 1990 147 219 6 L8 OLI 22 April 2015 112 13 MSS 2 August 1977 214 Table7 A2. L8Landsat OLI 7/8 (ETM+/OLI) 9 June 2015 acquisitions 160 of 14 Hispar TM glacier used 7 August to derive 1990 the velocity 219 fields depicted in Figure3 (as indicated) and Figure4 (all pairs). The time between two acquisitions is ∆t.

Date 1 Date 2 ∆t Sensor Figure Date 1 Date 2 ∆t Sensor Figure 5 July 2013 21 July 2013 16 L8 OLI 3a 29 January 2014 3 April 2014 64 L8 OLI Table21 July A2. 2013 Landsat 7 September 7/8 (ETM+/OLI) 2013 48 acquisitions L8 OLI 3b of Hispar 3 April 2014glacier used 6 June to 2014 derive 64the velocity L8 OLI fields depicted7 September in 2013 Figure 9 October3 (as indicated) 2013 32 and L8 Figure OLI 4 (all pairs). 6 June 2014The time between 8 July 2014 two acquisitions 32 L8 OLI is ∆t. 9 October 2013 25 October 2013 16 L8 OLI 8 July 2014 24 July 2014 16 L8 OLI 3c 25 October 2013 10 November 2013 16 L8 OLI 24 July 2014 25 August 2014 32 L8 OLI 3d Date 1 Date 2 ∆t Sensor Figure Date 1 Date 2 ∆t Sensor Figure 10 November 2013 26 November 2013 16 L8 OLI 25 August 2014 5 November 2014 16 L8 OLI 3e 265 November July 2013 2013 12 December 21 July 2013 2013 1616 L8 L8 OLI OLI 3a 295 November January 20142014 8 January 3 April 2015 2014 64 64 L7 L8 ETM+ OLI 3f/g 1221 December July 2013 2013 7 29 September January 2014 2013 48 48 L8 L8 OLI OLI 3b 8 3 JanuaryApril 2014 2015 9 February 6 June 20152014 3264 L7L8 ETM+ OLI 3h 7 September 2013 9 October 2013 32 L8 OLI 6 June 2014 8 July 2014 32 L8 OLI 9 October 2013 25 October 2013 16 L8 OLI 8 July 2014 24 July 2014 16 L8 OLI 3c 25 October 2013 10 November 2013 16 L8 OLI 24 July 2014 25 August 2014 32 L8 OLI 3d 10 November 2013 26 November 2013 16 L8 OLI 25 August 2014 05 November 2014 16 L8 OLI 3e 26 November 2013 12 December 2013 16 L8 OLI 5 November 2014 8 January 2015 64 L7 ETM+ 3f/g 12 December 2013 29 January 2014 48 L8 OLI 8 January 2015 9 February 2015 32 L7 ETM+ 3h

Remote Sens. 2017, 9, 888 17 of 18

Table A3. Image pairs from Landsat 7/8 (L7/L8) and Sentinel-1 (S1) descending orbits used to calculate velocities in the main surge phase (Figures5 and6). The time between two acquisitions (in days) is ∆t. The eight scenes marked in italics are used for Figure5.

Date 1 Date 2 ∆t Sensor Date 1 Date 2 ∆t Sensor 13 March 15 6 April 15 24 Sentinel-1 18 December 15 3 January 16 16 L8 OLI 6 April 15 18 April 15 12 Sentinel-1 3 January 16 19 January 16 16 L8 OLI 30 April 15 12 May 15 12 Sentinel-1 19 January 16 12 February 16 24 Sentinel-1 24 May 15 5 June 15 12 Sentinel-1 12 February 16 7 March 16 24 Sentinel-1 17 June 15 3 July 15 16 L7 ETM+ 7 March 16 31 March 16 24 Sentinel-1 23 July 15 16 August 15 24 Sentinel-1 31 March.16 24 April 16 24 Sentinel-1 16 August 15 9 September 15 24 Sentinel-1 24 April 16 18 May 16 24 Sentinel-1 13 September 15 15 October 15 32 L8 OLI 18 May 16 11 June 16 24 Sentinel-1 15 October 15 31 October 15 16 L8 OLI 11 June 16 29 July 16 48 Sentinel-1 31 October 15 2 December 15 32 L8 OLI 29 July 16 22 August 16 24 Sentinel-1 2 December 15 18 December 15 16 L8 OLI 22 August 16 15 September 16 24 Sentinel-1

References

1. Copland, L.; Sylvestre, T.; Bishop, M.P.; Shroder, J.F.; Seong, Y.B.; Owen, L.A.; Bush, A.; Kamp, U. Expanded and recently increased glacier surging in the Karakoram. Arct. Antarct. Alp. Res. 2011, 43, 503–516. [CrossRef] 2. Hewitt, K. Tributary glacier surges: An exceptional concentration at Panmah Glacier, Karakoram Himalaya. J. Glaciol. 2007, 53, 181–188. [CrossRef] 3. Rankl, M.; Kienholz, C.; Braun, M. Glacier changes in the Karakoram region mapped by multimission satellite imagery. Cryosphere 2014, 8, 977–989. [CrossRef] 4. Paul, F. Revealing glacier flow and surge dynamics from animated satellite image sequences: Examples from the Karakoram. Cryosphere 2015, 9, 2201–2214. [CrossRef] 5. Quincey, D.J.; Glasser, N.F.; Cook, S.J.; Luckman, A. Heterogeneity in Karakoram glacier surges. J. Geophys. Res. Earth Surf. 2015, 120.[CrossRef] 6. Rankl, M.; Braun, M. Glacier elevation and mass changes over the central Karakoram region estimated from TanDEM-X and SRTM/X-SAR digital elevation models. Ann. Glaciol. 2016, 51, 273–281. [CrossRef] 7. Jiskoot, H. Glacier surging. In Encyclopedia of Snow, Ice and Glaciers; Singh, V.P., Singh, P., Haritashya, U.K., Eds.; Springer: Heidelberg, Germany, 2011; pp. 415–428. 8. Kotlyakov, V.M.; Osipova, G.B.; Tsvetkov, D.G.; Jacka, J. Monitoring surging glaciers of the Pamirs, Central Asia, from space. Ann. Glaciol. 2008, 48, 125–134. [CrossRef] 9. Grant, K.L.; Stokes, C.R.; Evans, I.S. Identification and characteristics of surge-type glaciers on Novaya Zemlya, Russian Arctic. J. Glaciol. 2009, 55, 960–972. [CrossRef] 10. Herreid, S.; Truffer, M. Automated detection of unstable glacier flow and a spectrum of speedup behavior in the Alaska Range. J. Geophys. Res. Earth Surf. 2016, 121, 64–81. [CrossRef] 11. Copland, L.; Pope, S.; Bishop, M.P.; Shroder, J.F.; Clendon, P.; Bush, A.; Kamp, U.; Seong, Y.B.; Owen, L.A. Glacier velocities across the central Karakoram. Ann. Glaciol. 2009, 50, 41–49. [CrossRef] 12. Quincey, D.J.; Braun, M.; Glasser, N.F.; Bishop, M.P.; Hewitt, K.; Luckman, A. Karakoram glacier surge dynamics. Geophys. Res. Lett. 2011, 38.[CrossRef] 13. Bhambri, R.; Bolch, T.; Kawishwar, P.; Dobhal, D.P.; Srivastava, D.; Pratap, B. Heterogeneity in glacier response in the upper Shyok valley, northeast Karakoram. Cryosphere 2013, 7, 1385–1398. [CrossRef] 14. Gardelle, J.; Berthier, E.; Arnaud, Y.; Kääb, A. Region-wide glacier mass balances over the Pamir-Karakoram-Himalaya during 1999–2011. Cryosphere 2013, 7, 1263–1286. [CrossRef] 15. Eisen, O.; Harrison, W.D.; Raymond, C.F.; Echelmeyer, K.A.; Bender, G.A.; Gorda, J.L.D. Variegated Glacier, Alaska, USA: A century of surges. J. Glaciol. 2005, 51, 399–406. [CrossRef] 16. Sevestre, H.; Benn, D.I. Climatic and geometric controls on the global distribution of surge-type glaciers: Implications for a unifying model of surging. J. Glaciol. 2015, 61, 646–659. [CrossRef] 17. Quincey, D.J.; Luckman, A. Brief Communication: On the magnitude and frequency of Khurdopin glacier surge events. Cryosphere 2014, 8, 571–574. [CrossRef] 18. Fowler, H.J.; Archer, D.R. Conflicting signals of climatic change in the Upper Indus Basin. J. Clim. 2006, 19, 4276–4293. [CrossRef] Remote Sens. 2017, 9, 888 18 of 18

19. Bocchiola, D.; Diolaiuti, G. Recent (1980–2009) evidence of climate change in the upper Karakoram, Pakistan. Theor. Appl. Climatol. 2013, 113, 611–641. [CrossRef] 20. Bookhagen, B.; Burbank, D.W. Toward a complete Himalayan hydrological budget: Spatiotemporal distribution of snowmelt and rainfall and their impact on river discharge. J. Geophys. Res. 2010, 115.[CrossRef] 21. Dahri, Z.H.; Ludwig, F.; Moors, E.; Ahmad, B.; Khan, A.; Kabat, P. An appraisal of precipitation distribution in the high-altitude catchments of the Indus basin. Sci. Total Environ. 2016, 548–549, 289–306. [CrossRef] 22. Wake, C.P. Glaciochemical investigations as a tool to determine the spatial variation of snow accumulation in the Central Karakoram, Northern Pakistan. Ann. Glaciol. 1989, 13, 279–284. [CrossRef] 23. Hewitt, K. Glacier Change, Concentration, and Elevation Effects in the Karakoram Himalaya, Upper Indus Basin. Mt. Res. Dev. 2011, 31, 188–200. [CrossRef] 24. Kääb, A.; Treichler, D.; Nuth, C.; Berthier, E. Brief Communication: Contending estimates of 2003–2008 glacier mass balance over the Pamir-Karakoram-Himalaya. Cryosphere 2015, 9, 557–564. [CrossRef] 25. Pfeffer, W.T.; Arendt, A.A.; Bliss, A.; Bolch, T.; Cogley, J.G.; Gardner, A.S.; Hagen, J.O.; Hock, R.; Kaser, G.; Kienholz, C.; et al. The Randolph Glacier Inventory: A globally complete inventory of glaciers. J. Glaciol. 2014, 60, 537–552. [CrossRef] 26. Kääb, A.; Vollmer, M. Surface geometry, thickness changes and flow fields on creeping mountain permafrost: Automatic extraction by digital image analysis. Permafr. Periglac. Proc. 2000, 11, 315–326. [CrossRef] 27. Heid, T.; Kääb, A. Evaluation of different image matching methods for deriving glacier surface displacements globally from optical satellite imagery. Remote Sens. Environ. 2012, 118, 339–355. [CrossRef] 28. Strozzi, T.; Murray, T.; Wegmüller, U.; Werner, C. Glacier Motion Estimation Using SAR Offset-Tracking Procedures. IEEE Trans. Geosci. Remote Sens. 2002, 40, 2384–2391. [CrossRef] 29. Wake, C.; Searle, M. Rapid advance of Pumarikish Glacier, Hispar Glacier Basin, Karakoram Himalaya. J. Glaciol. 1993, 39, 204–206. [CrossRef] 30. Strozzi, T.; Kääb, A.; Schellenberger, T. Frontal destabilization of Stonebreen, Edgeøya, Svalbard. Cryosphere 2017, 11, 553–566. [CrossRef] 31. Frezzotti, M.; Capra, A.; Vittuari, L. Comparison between glacier velocities inferred from GPS and sequential satellite images. Ann. Glaciol. 1998, 27, 54–60. 32. Dehecq, A.; Gourmelen, N.; Trouve, E. Deriving large-scale glacier velocities from a complete satellite archive: Application to the Pamir-Karakoram-Himalaya. Remote Sens. Environ. 2015, 162, 55–56. [CrossRef] 33. Wilson, R.; Mernild, S.H.; Malmros, J.K.; Bravo, C.; Carrión, D. Surface velocity fluctuations for Glaciar Universidad, central Chile, between 1967 and 2015. J. Glaciol. 2016, 62, 847–860. [CrossRef] 34. Kääb, A.; Winsvold, S.H.; Altena, B.; Nuth, C.; Nagler, T.; Wuite, J. Glacier Remote Sensing Using Sentinel-2. Part I: Radiometric and Geometric Performance, and Application to Ice Velocity. Remote Sens. 2016, 8, 598. [CrossRef] 35. Raymond, C.F. How do glaciers surge? A review. J. Geophys. Res. 1987, 92, 9121–9134. [CrossRef] 36. Dunse, T.; Schellenberger, T.; Hagen, J.O.; Kääb, A.; Schuler, T.V.; Reijmer, C.H. Glacier-surge mechanisms promoted by a hydro-thermodynamic feedback to summer melt. Cryosphere 2015, 9, 197–215. [CrossRef] 37. Round, V.; Leinss, S.; Huss, M.; Haemmig, C.; Hajnsek, I. Surge dynamics and lake outbursts of Kyagar Glacier, Karakoram. Cryosphere 2017, 11, 723–739. [CrossRef] 38. Yasuda, T.; Furuya, M. Dynamics of surge-type glaciers in West Kunlun Shan, Northwestern Tibet. J. Geophys. Res. Earth Surf. 2015, 120, 2393–2405. [CrossRef] 39. Bolch, T.; Pieczonka, T.; Mukherjee, K.; Shea, J. Brief communication: Glaciers in the Hunza catchment (Karakoram) have been nearly in balance since the 1970s. Cryosphere 2017, 11, 531–539. [CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).