Although animated images are very popular on the internet, they have so far
found only limited use for glaciological applications. With long time series
of satellite images becoming increasingly available and glaciers being well
recognized for their rapid changes and variable flow dynamics, animated
sequences of multiple satellite images reveal glacier dynamics in a
time-lapse mode, making the otherwise slow changes of glacier movement
visible and understandable to the wider public. For this study, animated
image sequences were created for four regions in the central Karakoram
mountain range over a 25-year time period (1990–2015) from freely available
image quick-looks of orthorectified Landsat scenes. The animations play
automatically in a web browser and reveal highly complex patterns of glacier
flow and surge dynamics that are difficult to obtain by other methods. In
contrast to other regions, surging glaciers in the Karakoram are often small
(10
Analysis of sequential satellite images has become a common tool for deriving glacier changes through time in all parts of the world. A “standard” way of documenting these changes in scientific journals is the overlay of glacier outlines from different points in time on one of the images used for the analysis (e.g. Baumann et al., 2009; Bhambri et al., 2013; Paul et al., 2004). In the case where multiple images are available and changes take place mostly at the glacier terminus (e.g. during an advance or retreat phase), terminus positions are indicated by multiple lines with years either attached to them (e.g. Jiskoot and Juhlin, 2009) or colour-coded (McNabb and Hock, 2014; Quincey et al., 2011; Rankl et al., 2014). When complex interactions take place between two glaciers (e.g. a tributary is merging with another glacier), phases of the changes are illustrated by showing sequential images side by side (e.g. Belò et al., 2008; Bhambri et al., 2013; Copland et al., 2011; Mukhopadhyay and Khan, 2014) or by two-dimensional drawings of changes in major moraine patterns (e.g. Hewitt, 2007; Meier and Post, 1969; Quincey et al., 2015).
Although these representations of changing glaciers are scientifically sound
and exact, they have some limitations in demonstrating dynamic aspects. The
key issue is related to the limited ability of the human brain to recognize
differences between two (static) images when shown side by side or to
translate various outlines of terminus positions into the correct sequence of
changes, in particular when changes are out of phase for a couple of
glaciers. On the other hand, the human brain recognizes movement well and
tends to compensate missing parts in a sequence of animated images due to the
slow processing of visual information, also known as the “phi phenomenon”
(e.g. MacGillivray, 2007). This helps in translating time-lapse photography
into continuous motion, thus making the dynamic nature of otherwise slowly
moving objects or natural phenomena visible (e.g. cloud development, aurora,
tides). While cameras with an interval timer were not common a decade ago and
related footage was rare, today's widespread availability of webcams allows
pictures to be taken remotely and automatically each day (or whatever period)
at regular intervals. This can be particularly interesting when glaciers are
imaged, as their movement is normally much too slow to be recognized (e.g.
At the satellite scale, the application of “flicker” images (basically a
rapid alternation of two images taken a few years apart) for demonstrating
glacier changes is common practice and has been used to analyse glacier
motion (Kääb et al., 2003). In this way, coherent patterns of
displacement of the glacier surface have long been used to determine surface
flow velocities from feature tracking using cross-correlation or other
techniques (e.g. Kääb and Vollmer, 2000; Scambos et al., 1992; Paul
et al., 2015). With the now free availability of long time series (starting
in 1984) of orthorectified satellite imagery from Landsat (e.g. Wulder et
al., 2012), it is possible to combine sequential satellite images into longer
sequences (
In this study animated sequences of orthorectified satellite images covering
a 25-year time period (1990–2015) are used to demonstrate glacier dynamics
and other landscape changes in four regions of the central Karakoram. Though
this might be seen as a less quantitative approach than that of studies
determining the exact rates of glacier change, the information obtained by
looking at high-speed animations of the individual images also provides
insight into dynamic glacier behaviour on a different level. There is also
potential for using such animations for educational purposes by visualizing
how glaciers flow and change through time. The animations use the very old
(
The Karakoram mountain range has been selected due to its many surging
glaciers that display a distinct dynamic behaviour (e.g. Copland et
al., 2011; Gardelle et al., 2013; Hewitt, 2007; Rankl et al., 2014).
According to Jiskoot (2011), a surge-type glacier oscillates between a period
of slow or normal flow (lasting tens to hundreds of years) named the
While the Karakoram region is well known for its many surge-type glaciers (e.g. Copland et al., 2011; Hewitt, 2014), counting them is challenging as the frequently used criteria for their identification only partly apply. Many studies have thus introduced a “surge index” to indicate the certainty that a specific glacier is of surge type (cf. Sevestre and Benn, 2015). The evidence can be divided into geomorphological and dynamic categories (e.g. Jiskoot, 2011). The former include looped or distorted medial moraines, a glacier tongue that is largely covered by crevasses and séracs during a surge, a post-surge disconnection of the tongue well behind the terminus, and rapid downwasting after the surge with the formation of potholes and remaining stranded icebergs (e.g. Yde and Knudsen, 2005). Dynamic criteria include (among others) the terminus advance rate, the total advance over a given period, the duration of the advance and retreat (or quiescent) phase, the relative advance compared to the pre-advance glacier length, absolute values of surface velocity, significant velocity changes in specific regions of a glacier, surge periodicity and inverse thickness changes in the ablation (mass gain) and accumulation (mass loss) regions. For these dynamic criteria, the values for surging glaciers can be 1 to 3 orders of magnitude higher than for non-surge-type glaciers (e.g. Jiskoot, 2011). However, they can also lie within a similar range, thus limiting the possibilities for a clear separation. For this study a glacier is called “surging” based on its easily identifiable strong and partly rapid advance or on the basis of its classification in previous studies (Copland et al., 2011; Gardelle et al., 2013; Rankl et al., 2014). In the event that other criteria are used (e.g. large changes in flow velocity), a different assignment might result.
The study region is located in the central Karakoram mountain range (Fig. 1) to the north of – and including – the large and well-studied Baltoro Glacier (Quincey et al., 2009 and references therein). Four regions are selected for the animations: (1) Baltoro, (2) Panmah, (3) Skamri/Sarpo Laggo and (4) Shaksgam. All regions are well known for their many surge-type glaciers (cf. Copland et al., 2011; Rankl et al., 2014), several of which have been studied in more detail (Diolaiuti et al., 2003; Hewitt, 2007; Quincey et al., 2011; Rankl et al., 2014). The region is characterized by very steep and high terrain (often reaching more than 7000 m a.s.l.) with numerous multi-basin valley glaciers, that often have further tributary glaciers in their ablation zones (Iturrizaga, 2011). The anomalous glacier behaviour in the study region (mass gain and advancing glaciers over the past 2 decades), relative to most other regions of the world has been named the “Karakoram Anomaly” (e.g. Bolch et al., 2012; Hewitt, 2005). This behaviour might be attributable to an increase in precipitation (e.g. Janes and Bush, 2012), but the large number of actively surging glaciers in the region might also have non-climatic causes (e.g. Hewitt, 2005; Jiskoot, 2011). A recent study by Sevestre and Benn (2015) has suggested that glaciers in this region are located in the climatically “correct” zone for surge-type glaciers. Further details about the topo-climatic characteristics of the region can be found in Hewitt (2014).
Landsat scene of the study region from 2004 showing footprints of
the four subregions depicted in Figs. 2–5. The black square in the inset
shows the location of the study region in the Karakoram mountain range (map
taken from Google Earth). The image centre is at 36
The study region is completely covered by Landsat scene 148–35 (path–row)
and partly by scene 149–35 (Fig. 1). Useful Landsat scenes (sensors TM,
ETM
Overview of the 21 Landsat scenes used to create the animations in
the Supplement for the four subregions shown in Figs. 2–5. The MSS scene is
only added for completeness. Abbreviations are as follows. From the
“sensor” column, MSS: MultiSpectral Scanner, TM: Thematic Mapper, ETM
The animations are created by displaying all images in a GIS (e.g. QGIS,
ArcMap), exporting the maps to a 24-bit image file, converting all images to
GIF format and by creating the animated GIF image with a delay of 1/10
A wide variety of dynamic changes are visible in the animations. They range from the variable extent of seasonal snow cover, to clouds (with their shadows) that pop up on individual frames, to the steady flow of large glaciers with traceable surface characteristics (moraines, lakes), to glacier fronts advancing and retreating at different rates, mass waves travelling through individual glaciers, downwasting of debris-covered glacier sections and short-term velocity pulses within a glacier. As all of the above are happening at the same time, it is easy to lose focus. Following the evolution of specific changes thus requires focussing the view on a specific region and ignoring everything else. In the following, some prominent observations are described for each subregion.
In subregion (1) Baltoro Glacier and its numerous tributaries dominate the scene (Fig. 2). With few exceptions, all glaciers show continuous and near-steady flow. Despite the easily recognizable high flow velocity of the main glacier, its terminus remains in about the same position. Supraglacial lakes across its surface appear and disappear but are easily traceable markers for the surface flow. Towards the upper parts of the glacier (to the right of the image), flow velocity seems to increase, because tracing features becomes increasingly difficult. In the accumulation region, flow dynamics are difficult to follow due to a lack of traceable features and the high variability of snow extent. However, in the lower right of the image, steep snow-covered glacier headwalls (facing north) and some deep crevasses reveal very high flow velocities. The flow speed here is high enough that the 1–3-year time step between images fails to provide the impression of continuous flow; instead it more looks like a nervous shaking.
Subregion (1) (Baltoro) shows the tongue of Baltoro Glacier and its surrounding tributaries. SG (orange): actively surging glacier, SG (white): surge-type glacier (only on Figs. 4 and 5), A: advancing glacier. The Landsat scene 148-035 is from 14 August 2004.
As the animations are particularly useful for recognizing changes, the advancing fronts of several rather small glaciers with narrow and/or heavily debris-covered tongues are also easily visible. In most cases it would have been nearly impossible to spot these changes from comparison of image pairs alone. This is also the case for several advancing glaciers in the lower part of the image. Two surge-type glaciers (Trango and Muztagh in Fig. 2) with distorted moraines and a clearly visible variability in surface elevation can be seen to the north of Baltoro. To the south of them is the surge-type Liligo Glacier, which has been studied in detail before (e.g. Belò et al., 2008). This glacier reached its maximum extent around 1998 and was in its downwasting or quiescent phase afterwards. As the animations cover the surge and post-surge periods, the rapid advance of the front during the surge and the following thinning and disconnection of the lower tongue can be traced very well. Three of the four rather small glaciers in the southwest corner of the image (marked with an “x” in Fig. 2) show a behaviour that is similar to Liligo, in particular the one to the very left. At the same time, the terminus of the glacier to its right goes slightly back and forth several times.
In the Panmah region (Fig. 3, subregion (2)), the variability in late summer snow extent, and the differences between the behaviour of the steady-flowing and actively surging glaciers are easily recognizable. While the large tongues of Biafo, Choktoi and Nobande Sobonde (NS) glaciers show the steady flow typical of non-surge-type glaciers, several (partly tributary) glaciers show strong advances (First Feriole, Shingchukpi), sudden onset of fast flow (Drenmang), or mass waves travelling down glacier (Chiring), and most of them finally collide with other glaciers (e.g. Maedan, Chiring) and create the well-known distorted and looped moraines (Hewitt, 2007). Also apparent is the asynchronous nature of the advance/retreat (or downwasting) phases. While Chiring Glacier finished its surge before 1998 (cf. Hewitt, 1998), other glaciers either started to surge around 2000 (First Feriole), were already in full surge mode then (e.g. Shingchukpi), or began to surge later (e.g. in 2006 for Drenmang). The surge of First Feriole glacier started while the ice masses from the previous surge were still down and backwasting in the valley floor. This gives the impression of one retreating and one advancing terminus at the same time. As in the Baltoro region, the termini of many of the much smaller surrounding glaciers are either stationary or advancing slowly. Some of the advancing glaciers have a terminus width of only 1 or 2 pixels but their changes can be easily followed in the animation.
Subregion (2) (Panmah) shows the region around Panmah and Choktoi glaciers with surrounding tributaries; annotations and Landsat scene as in Fig. 2.
Subregion (3) (Skamri) shows the region between Skamri and Sarpo Laggo glaciers; annotations and Landsat scene as in Fig. 2.
The variability described above for subregion (2) is also apparent in subregion (3), as depicted in Fig. 4. Many of the surge-type glaciers are actively surging during the period of observation and only a few are in their quiescent phase. The smaller non-surge-type glaciers are either stagnant or slightly advancing, while the large debris-covered glaciers are stagnant and downwasting. A wide range of terminus advance rates is apparent as well. While one glacier (North Chongtar) advances very slowly (and might not be identified as being surge-type from its advance rate), others advance rapidly and strongly. They partly merge for some time with a larger main glacier (Sarpo Laggo and Skamri). As some of these glaciers (marked with an “x” in Fig. 4) retreat back to their former positions over the remainder of the time period, the looping of the animations creates the impression that they are pulsating. This is different for South Chongtar Glacier, which has a downwasting and retreating ice mass from a previous surge in front of its more or less stagnant terminus. However, on closer inspection, one can see a surge front, a somewhat wider and advancing region up-glacier that is travelling down towards the stagnant terminus, likely indicating a forthcoming surge. If this is already termed a surge, then this glacier is in its quiescent and surge phase at the same time (demonstrating the difficulties for using the terminology correctly at a given time).
The small glacier on the opposite side of the valley is also advancing and might again (as in 1977) surge down to the tongue of Sarpo Laggo Glacier. Moni Glacier (Fig. 7) shows some interesting interaction with Sarpo Laggo. It seems that the lobate tongue of Moni (resulting from a previous surge) blocked the flow from Sarpo Laggo for some time and a fast flow event within the blocked ice mass advected the lobate structure from Moni Glacier quickly down valley for some years, as has also been observed for several surging glaciers in the Pamir (Kotlyakov et al., 2008). During and towards the end of this event, a substantial thinning of Sarpo Laggo can be observed upstream. From the moraine deformation visible in the last images, it also seems that Moni Glacier was again pushing into Sarpo Laggo, maybe starting its next surge. A similar interaction might also happen for South Skamri and Skamri glaciers. The animations reveal the increasing blocking of the flow of the main glacier by the surge of its tributary, South Skamri Glacier. So far, Skamri Glacier has not reacted to this blocking, but a sudden push of the distorted moraine in the upper left is already visible (see also Fig. 3c in Quincey et al., 2015). It remains to be seen if this push will catch up with the still ongoing surge of South Skamri.
In subregion (4) such a short-lived high-speed push event can be seen on Crown Glacier. The most massive surge can be seen for North Crown Glacier that has also reactivated the downwasting ice masses from a previous surge of a similar-sized neighbouring glacier in its own surge (cf. Fig. 8 in Quincey et al., 2015). This glacier also shows a mass wave that is travelling down to the front creating the strong advance. Several other larger glaciers in the region are also advancing or even surging, depending on the definition of a surge. The ongoing surge of Tatulu Guo Glacier, shown in the upper right of Fig. 5, has also been described by Quincey et al. (2011). Several smaller glaciers show either stationary or slowly advancing fronts.
As mentioned above, changes in surface elevation can be followed in the animations, in particular where there are stable lateral moraines. Changes observed include the thickening/thinning pattern typical of surging glaciers that occurs as a surge front propagates down glacier (as for Chiring, Drenmang and North Crown glaciers), as well as the downwasting of the stagnant tongues of several of the large debris-covered glaciers, in particular in subregion (3) (Sarpo Laggo, Skamri). There are also glaciers where no elevation changes can be detected, such as Baltoro and Choktoi glaciers, where flow appears to be stable. Rather interesting is also the surface lowering of Panmah Glacier (the lower part of Nobande Sobonde) in subregion (2), which occurs despite additional mass input from surging tributaries (Chiring and Shingchukpi). More subtle are the elevation increases and decreases that result from flow blocking such as that which characterizes the interaction between Moni and Sarpo Laggo or Drenmang and NS glaciers. In particular the latter can also be seen in the study by Gardelle et al. (2013), who determined elevation changes over the 2000–2008 period using differencing of two digital elevation models (DEMs). The animations reveal how these changes took place and how they were related to other changes associated with a glacier surge, such as frontal advance/retreat and velocity changes.
Subregion (4) (Shaksgam) shows the region to the north of Skamri Glacier to both sides of the Shaksgam Valley; annotations and Landsat scene as in Fig. 2.
Another form of variability can be seen in the numerous (hundreds) supraglacial lakes and ponds covering the lower parts of Baltoro Glacier (subregion (1)), Panmah (subregion (2)) and some other glaciers. These lakes seem to be rather short-lived (about 2–3 years), limiting their use for determining flow velocities by feature tracking to a 1-year period. Most of the lakes are about the same size but their shape varies rather strongly from scene to scene. For Baltoro Glacier, it is apparent that supraglacial lakes often form in zones of compressive flow (where larger tributaries join), indicating that surface meltwater is not efficiently drained. However, it also has to be considered that the images are taken at different dates each year and that the extent and level of the lakes varies over a year in response to changing melt conditions and rainfall. Stationary lakes outside the lateral moraines also change size over time.
Supraglacial debris occurs in form of the typical flow-parallel moraines (where they first appear) that often spread out to create a complete debris cover near the terminus, in the form of distorted (wave-like) bands where glaciers with unsteady flow interact, or in the form of local debris accumulations from rock fall events. One glacier (Mundu) in subregion (1) has regular and similar-sized patches of debris on its surface, indicating periodic rock fall activity.
Finally, subregion (2) shows local movement of terrain that is actually stable, mostly along mountain peaks and ridges. This is likely the result of the use of different DEMs to orthorectify the satellite images. As this apparent movement is concentrated on regions outside the glaciers (i.e. on “stable” terrain), an algorithm for calculating flow velocity would obtain a considerable surface displacement in these regions, which has to be removed manually before assessment of the accuracy of glacier velocities over stable terrain can be performed. The animated images clearly reveal such regions, helping to determine the quality of the orthorectification for an entire time series and identify the obviously unstable regions for the post-processing of velocity data (e.g. Kääb, 2005).
The animations presented here would not have been possible without the
accurate and consistent orthorectification of all satellite scenes by USGS.
Indeed, errors in the DEM used for orthorectification translate into
incorrect pixel positions that change when a different DEM is used for
orthorectification, especially in high relief and steep terrain (see
Sect. 3.4). However, for relatively flat glacier tongues, this effect is
small and does not obscure the much larger changes due to glacier flow. It
must also be mentioned that the animations are created from image quick-looks
that are freely provided (
With the processing line being established, scenes must have been acquired
over a sufficiently long time period (at best each year) and be available in
the USGS archive. The former point is not necessarily the case for the first
years of Landsat operation and during its commercial phase (see Goward et
al., 2006). The latter point (transfer of scenes from non-USGS archives to
USGS) is an ongoing process that will continuously expand the possibilities
for other regions. Data availability is also restricted after 2003, when the
scan-line corrector of ETM
On the more practical side, it is important that appropriate scenes are
available. Key requirements are that all scenes are taken in each year near
the end of the ablation period (with as much glacier ice exposed as possible)
at about the same day (to avoid strong changes due to shadow) and without
clouds. While it was not possible to have identical extents of snow cover on
all images used in this study (resulting in a rather nervous flickering when
animated), the variability of shadow extent is rather limited due to the high
solar elevation. Clouds were an issue in some years and regions (e.g. in
2008), but the large data gap between 2004 and 2009 is mostly due to missing
stripe-filled quick-looks as suitable ETM
The final step in creating the images was the selection of the most useful
scenes for each subregion. This was a compromise between (a) using as many
largely cloud-free images as possible, (b) having a more or less constant
time step of 1–3 years between each image, (c) avoiding strong changes in
reflectivity due to seasonal snow, and (d) capturing the relevant processes
that sometimes occur only in a specific year. As Table 1 reveals, it was not
really possible to satisfy point (b) for any of the regions without
compromising the other points. This results in a non-equidistant temporal
difference between the images (from 1 to 5 years) with potential effects on
the perception of flow velocities (e.g. flow seems faster with 3-year than
with 1-year time steps). Obvious changes in flow velocity have thus to be
interpreted with great care. As a compromise, for some regions, more
individual images are provided on the separate server
(
Overall, the animations presented here provide an overview on glacier and landscape dynamics over a period of 25 years that cannot be created in all regions of the world. Apart from the above criteria related to the availability and selection of images, glaciers also have to be large and structured enough to reveal flow dynamics. Otherwise, it will be mostly the changes in extent that can be followed, but this might be of interest as well.
As described in the introduction, several possibilities exist to visualize glacier changes (e.g. front variations) or dynamics (e.g. flow fields and their changes over a given time period) in a quantitative way (e.g. overlay of colour-coded and/or annotated outlines or maps, side by side comparisons). The animations on the other hand, provide information and insights that are complementary to the above classic (quantitative) ways. Apart from the fact that the animations reveal the temporal evolution of otherwise very slow processes in a much faster (time-lapse) mode and provide a true dynamic feeling, they also give a holistic view of changes taking place at the same time (e.g. terminus fluctuations, elevation changes, flow velocities) that is nearly impossible to obtain from static maps. Examples are the stagnant downwasting lower tongues of Skamri and Sarpo Laggo glaciers, and the mass wave that is travelling down Chiring Glacier at high velocity, resulting in a fast and strong advance of the terminus. This easily visible combination of effects might ultimately help to establish a more comprehensive definition of surge-type or surging glaciers that seemingly needs to be expanded to correctly capture the full variability of possible glacier dynamics.
When looking at the benefits of animations more systematically, they can be
roughly grouped into three categories: (I) more subtle insights facilitating
process understanding, (II) new scientific information, and (III) technical
advantages. Insights of category (I) are
how a glacier flows in general and how a main stream of ice is fed by
its tributaries; the succession or timing of events for individual glaciers and the
entire region; understanding of glacier dynamics for the wider public that is not
used to interpret colour-coded velocity fields or line graphs.
Insights of category (II) include
identification of surging glaciers using a range of criteria that are
all visible at the same time (e.g. advance rates of the terminus, internal
velocity changes, travelling of mass waves through the glacier, typical
downwasting pattern after a surge); for long time series, information retrieval is maximized when full
surge cycles are captured; a holistic and at the same time very detailed view of decadal glacier
variability and dynamics in a large region; the gradual downwasting of mostly large, stationary and debris-covered
glaciers can be followed through time; the very high flow velocities of the ice in steep parts of accumulation
regions that are not captured by traditional methods such as offset tracking
from optical (contrast issue, stationary crevasses) or microwave data (radar
shadow, decorrelation); the rapid changes (formation and decay) of supraglacial lakes; the identification of very tiny (1–2 pixel) debris-covered glacier
fronts (that are hard to recognize in individual images) and their changes
through time; movement of terrain that should be stable but is not, due to different
DEMs being used for orthorectification (important for accuracy assessment of
velocity products).
In principle, most of the above points can also be detected if individual
images are uploaded in an image browser by scrolling through them back and
forth. So what are the technical advantages (category III) of using the
automated animations?
On the practical side, the manual scrolling through the time series can
only be performed a few dozen times before either the finger becomes tired or the scroll wheel stops working; similarly, an automated animation allows the mouse to be used to point
out specific details; more technically, for large data sets (e.g. more than six images) it
might not be possible to include all images in the series within a single
scroll; scientifically, the constant repetition rate of the individual images
cannot be achieved by manual scrolling (important to recognize changes in
flow velocity); visually, the back and forth movement of the ice is disruptive and
does not provide the same category (I) insights (e.g. of continuous flow); finally, the repetition rates can be adjusted (lowered) to study
specific changes in detail.
Of course, animated flicker images provide similar advantages, but they are normally restricted to two images and do thus not show the full temporal development over a longer time. They also suffer from the back and forth effect (IIIe) that hinders the impression of continuous flow. However, they can be very practical for rapid change detection when image conditions are about the same.
As mentioned above, there is certainly potential for creating or using
animations for educational purposes. While the visualization of glacier flow
dynamics itself might be of interest for teaching, public communication or
exhibitions, classroom experiments might look at using other images for each
time series (fewer/more), changing the repetition rate, analyse effects of
the looping, and adding annotations such as a time bar.
Remote-sensing-related questions might focus on the spectral properties of
ice and snow and the false colour composites used, spatial resolution and
visibility of details, or the value of long-term time series and free data
availability. The latter might be further explored in hands-on lectures or
summer schools (e.g. Manakos et al., 2007) by creating such animations for
other regions with sufficient temporal coverage. The animations might also
help understanding of natural variability over timescales that are not
available from any other source. Finally, the time-lapse mode compresses a
25-year development into 1
Due to the lack of contrast and traceable features, optical sensors usually fail to provide information on flow velocities in the accumulation regions of glaciers (e.g. Quincey et al., 2009; Dehecq et al., 2015). The same is true for microwave sensors in steep terrain due to radar shadow and layover effects (e.g. Rott, 2009). However, high-resolution missions such as TerraSAR-X have improved the situation to some extent (Rankl et al., 2014) and show localized regions of fast flow, even at the highest elevations of glaciers. The animations reveal that the regions of fast flow are rather widespread in steep accumulation regions and not adequately captured by current velocity maps.
As mentioned in the study by Sevestre and Benn (2015) and several previous investigations (e.g. Copland et al., 2011; Hewitt, 2007; Quincey et al., 2011; Rankl et al., 2014), the central Karakoram has a high abundance of surge-type glaciers of which many have actively surged in the past 25 years. As “normal” glacier advances are basically a consequence of changed climatic conditions, while surges largely result from internal mechanisms (e.g. Jiskoot, 2011; Meier and Post, 1969; Raymond, 1987; Sharp, 1988), it is important to distinguish the two glacier types. However, the animations reveal a large heterogeneity of the surging glaciers in terms of size, hypsometry, exposure, advance rates, surge durations, etc. that clearly overlap with the characteristics of non-surge-type glaciers.
While the frontal advance rate, duration or distance are only three of many
criteria used to identify glaciers as surging (see overview in Sevestre and
Benn, 2015), the above examples reveal that there is actually a continuum of
advance rates that allows no clear separation between surging and advancing
glaciers. The same is true for advance durations that vary from short pulses
(1–2 years) of rapid advance (Drenmang) to slow advances taking more than 10
or even 25 years (First Feriole, North Chongtar). Similar advance rates and durations can
also be found for non-surge-type glaciers. Moreover, glaciers of nearly any
size seem to surge, from small (
As previous studies have shown (e.g. Raymond, 1987), some glacier surges
involve considerable changes in flow velocity, surface elevation and extent.
Such quasi-parallel changes are recognizable in the animations for some
glaciers in subregion (2) (Chiring, Drenmang). However, in several cases,
only some of these changes occur and for this study the characterization is
based on easily recognizable advances, consistent with previous literature
(Copland et al., 2011; Rankl et al., 2014; Quincey et al., 2014) and
historic satellite images (MSS scene from 1977). However, glacier 14 in
Fig. 6 of the study by Rankl et al. (2014) is only marked as
Previous studies that have characterized surge-type glaciers according to
their topographic characteristics (e.g. area, length, slope, debris cover)
have found a tendency for surge-type glaciers to be longer, less steep, with
more branches, and being more fully debris-covered than non-surge-type
glaciers of similar size (e.g. Clarke et al., 1986; Barrand and Murray, 2006;
Rankl et al., 2014; Sevestre and Benn, 2015). In contrast, many of the
surge-type or surging glaciers in the study region are comparatively small
(2–20
Another possibility for separating surging from normal advancing/retreating
glaciers is related to their specific post-surge behaviour (i.e. the
quiescent phase). As the animations reveal (for glaciers that do not flow
into another glacier from the outset), the way the extended tongue
down-wastes and disintegrates after a surge is rather specific. It seems
(e.g. for Liligo in Fig. 2 or Shingchukpi in Fig. 3) that the entire surged
ice mass is transformed into dead ice after a surge and decays by
downwasting, similar to the ice resulting from a dry calving event. After
some years, this downwasting separates the lower part of the surged ice mass
from an upper part at about
Many of the glaciers in the study region have reportedly surged during the past century (cf. Copland et al., 2011), and historic satellite imagery (e.g. the MSS scene from 1977) reveals different extents of the surge-type glaciers analysed here. For example, First Feriole Glacier was in contact with Panmah Glacier in 1977 and the latest high-resolution satellite image (from 6 June 2014) available in Google Earth (Fig. 6) reveals that the glacier is still in full surge mode and might re-establish contact with Panmah Glacier in 2 or 3 years, resulting in a ca. 40-year surge cycle. A tributary of Sarpo Laggo in subregion (3) (no. 45 in Copland et al., 2011; no. 16 in Rankl et al., 2014) was in contact with the main glacier back in 1977, 1991 and again in 2007, indicating a ca. 15-year cycle. An image of its surge front from July 2006 is shown in Fig. 7. This is about 1.5 years before the glacier came into contact with Sarpo Laggo Glacier. Such a periodic repetition of surges is also a good indicator of surge-type behaviour.
The still-advancing (surging) tongue of First Feriole Glacier in the Panmah subregion. The image is a screenshot from Google Maps and was acquired on 6 June 2014.
A surging tributary of Sarpo Laggo Glacier can be seen to the right of
the centre of the image, as viewed from Moni Glacier in the foreground (see Sect. 3.1.3 for
details). To the left of the centre of the image, another unnamed surging glacier is
visible. The photo was taken in 2006 by Michael Beck (
As glacier surges might have non-climatic controls, the frontal changes of the non-surge-type glaciers can be better interpreted in climatic terms. In all four subregions, these glaciers have had either stable or advancing termini over the 25-year period (see Sect. 3.1). This implies that past mass budgets have generally been close to zero or even positive (e.g. Janes and Bush, 2012). Unfortunately, there is only indirect evidence for this as these glaciers are in general too small to obtain reliable geodetic mass budgets from DEM differencing (Gardelle et al., 2013). On the other hand, the characteristic mass loss at higher elevations and gain in the ablation region is easily visible for glaciers that surged after 2000 in the elevation difference grids presented by Gardelle et al. (2013). The resulting near-zero mass budgets of these glaciers are, however, only half the story. On a longer timescale, the surged ice masses also melt down and result in an overall mass loss of the glacier. For several glaciers the downwasting and mass loss after a surge can be traced in the animations. Downwasting also occurs in the ablation regions of some larger, debris-covered glaciers that are influenced by surges of tributaries (Panmah, Sarpo Laggo or Skamri). In this case it might be possible that the tributary surges move the ice of the main glacier to lower elevations where the debris cover is insufficient to protect the ice from melting. When considering these complex dynamic interactions, it seems appropriate to exclude surge-type glaciers from climate change impact studies that are related to timescales shorter than a full surge cycle.
This study discussed and presented (in the Supplement) animated
satellite image sequences from four regions in the central Karakoram mountain
range, covering a 25-year time period. The high repetition rate of
The animations also reveal variations that have not been reported before
(e.g. the high flow velocities in steep accumulation regions) and that would
be difficult to detect using other methods (e.g. advances of very small
debris-covered glaciers). The automated looping through all images has a
number of technical advantages over the routinely used flicker images (e.g.
full time series) and manual browsing (e.g. constant repetition rate) that
allow a more in-depth analysis of specific changes. There is great potential
for using the animations for educational purposes and further experimenting
with the combination of individual images (provided separately) and the frame
interval. Generating such animations for other glacierized regions would
likely provide additional insights, but it has to be considered that data
availability (e.g. due to clouds, snow cover, or unfilled ETM
As previous studies have shown, the study region is characterized by abundant surge-type glaciers, of which several were actively surging during the observation period 1990–2015. However, the animations reveal a wide spectrum of surge types from short-lived velocity pulses (without frontal advance), to mass waves travelling down-glacier (with strong frontal advance), to highly variable advance rates, distances and surge durations, all indicating that no single definition can be used to identify them as surge-type. It seems that several relatively small, steep and debris-free glaciers surge as well. As they lack the typical morphological evidence for surging, such as looped or distorted medial moraines, their surges can only be recognized from time-series analysis. Overall, the surges are generally out of phase with one another and some glaciers seem to surge periodically with repeat cycles of a few decades. The considerable overlap between the characteristics of surge-type and non-surge-type, but advancing, glaciers causes problems for classification.
The animations reveal that large glaciers with steady flow (Baltoro, Choktoi) have stationary terminus positions and stable surface elevations, while those influenced by surging tributaries (Panmah, Sarpo Laggo, Skamri) have stationary fronts but show considerable downwasting. In the latter case the ice of the main glacier might have been pushed downstream by a tributary surge. Considering such complex interactions and the possible non-climatic control of glacier surges, it seems advisable to exclude surge-type glaciers from the sample when climate change impacts are investigated on a timescale shorter than the surge cycle.
This study has been performed in the framework of the ESA project Glaciers_cci (4000109873/14/I-NB). All Landsat data were obtained from USGS. I would like to thank the scientific editor M. Sharp for his valuable comments and the careful review of the paper. Thanks are extended to B. Marzeion, D. Quincey and one anonymous referee for providing constructive reviews. S. Allen and P. Rastner gave helpful feedback to an earlier version of the paper.
Edited by: M. Sharp