3.1. Field measurements

In summer 2005, GPS measurements of surface motion were made at six locations on lower Baltoro glacier (Fig. 2). Locations were marked with a pole drilled into the ice, or with paint on large boulders, and their position surveyed with a Trimble R7 differential GPS. Occupation times at each location were ≥30 min, and the time between surveys varied between 14 and 22 days (Table 1). The data were differentially corrected using a semi-permanent base station in Skardu (∽80 km southwest of Baltoro glacier), temporary base stations at camp sites off the edges of the glacier, or the Precise Point Positioning (PPP) solution provided by Natural Resources Canada (http://www.geod.nrcan.gc.ca/products-produits/ppp_e.php). Positions are considered accurate to within ±0.05m horizontally and ±0.10m vertically. Note that these GPS measurements were collected prior to the 2006–07 period over which the feature-tracking measurements were made. However, they still provide the best available information concerning summer velocity patterns in the absence of any coincident field measurements from the feature-tracking period itself.

Fig. 2. Velocity patterns across Baltoro glacier and its northern tributaries derived from feature tracking. Velocity cross-profiles are shown for locations A–A' and B–Bllhe′. Numbered points indicate differential GPS measurement locations (Table 1). Arrows indicate location of feature-tracking match points, and calculated flow direction. Velocities not shown where <15ma^–1 or where feature tracking was not successful.

Table 1. Comparison between velocities derived from GPS vs feature-tracking (F-T) measurements on Baltoro glacier. All velocities are standardized to units of ma ^{_– 1}. The F-T velocities (from 2006–07) are provided from the closest match point to the GPS measurements (from summer 2005), within a maximum of four pixels (60 m) horizontal distance

3.2. Feature tracking

Surface velocities were derived from pairs of clear-sky satellite scenes by automated tracking of surface features such as crevasses and surface debris between scenes. The feature tracking was undertaken with VisiCORR Windows-based software (Reference DesioDowdeswell and Benham, 2003), which is based on the IMCORR image cross-correlation software developed by Reference RaymondScambos and others (1992). This method enables velocities to be determined to the sub-pixel level.

Georectified ASTER L1B scenes were used as input for the analyses, downloaded from NASA’s Earth Observing System Data Gateway. These images have a pixel size of 15 m in the visible and near-infrared bands used, and summer images were chosen to reduce the presence of snow cover. Images were also chosen based on their contrast and exposure quality, cloud cover, and acquisition characteristics; midday images from near-identical overlapping passes are most conducive to correlation processing. Ultimately, two pairs of scenes were used in the analysis: two scenes from adjacent acquisitions on 26 July 2006 (at 05:52:31 UTC and 05:52:40 UTC), and two scenes from adjacent acquisitions on 27 June 2007 (at 05:52:52 UTC and 05:53:01 UTC). Each scene covers a ground area of 60×60 km.

Initial processing involved image co-registration, followed by de-rotation of the images in ENVI software and extraction of band 3N in GeoTIFF format. Selected image sections, ranging from individual tributaries to full drainage basins, were then exported as eight-bit grayscale scenes and processed using VisiCORR. The main user-set parameters in VisiCORR are the search and reference chip sizes for the image cross-correlation calculations; after extensive testing, the optimal sizes were found to be 64 and 32 pixels, respectively. A grid spacing of eight pixels between adjacent correlation attempts was found to provide the best balance between processing speed and accurate velocity determinations. The final results from the feature-tracking calculations were output as a text file from VisiCORR, corrected to values of ma^–1, imported into ArcGIS software, and plotted on top of the original ASTER satellite imagery.

An assessment of the errors in the feature-tracking results was undertaken via an analysis of: (1) internal consistency of the velocity magnitudes (i.e. velocities should decrease in speed towards the glacier margins); (2) internal consistency of the velocity directions (i.e. glaciers should move in a generally downhill direction); and (3) the apparent movement of non-glaciated areas surrounding the glaciers (i.e. there should be no apparent motion in these regions). These checks indicated that the feature-tracking results generally produced realistic ice-motion patterns, with estimated errors in the velocity derivations being approximately ±1 pixel between scenes. As the ASTER scenes were acquired around 1 year apart, only ice motion >15ma^–1 is plotted here to remove any ambiguity in the results. In addition, velocities are not plotted where matches were not possible or where there were obvious mismatches that produced anomalously different velocities (in either magnitude or direction) from expected and/or from surrounding points.

3.3. Determination of ice-motion mechanisms

The relative importance of basal sliding vs internal deformation in accounting for observed surface motion can be determined from the spatial variability in velocities across transverse profiles. In areas where basal sliding dominates, ice tends to move en masse, with high but relatively constant velocities in the glacier center and rapid reductions close to the margins. This has been termed blockschollen (or plug flow) motion by some, particularly in relation to Karakoram glaciers (e.g. Reference Fatland, Lingle and TrufferFinsterwalder, 1937; Reference KääbKick, 1962), and is believed to represent a slab-like movement. By contrast, velocities that are generally low and increase gradually from the edges to the center of a glacier in a parabolic pattern are more likely to represent motion dominated by internal ice deformation. The surge of Variegated Glacier, Alaska, USA, provides an excellent example of the contrast between these motion types. Pre- and post-surge velocities are relatively low (∽0.10–0.20md^–1) and have a parabolic profile during periods dominated by deformational flow, whereas profiles at the same location during surges show high and constant motion (∽2.5–13md^–1) across almost the entire glacier when basal sliding is dominant (Kamb and others, 1985, their fig. 7). Transverse profiles have been used to infer the importance of basal sliding vs internal deformation from remote-sensing measurements on other glaciers (e.g. Reference Dowdeswell and BenhamFatland and others, 2003), and here we use them to make a general assessment of their relative importance on central Kara-koram glaciers.

4.1. Baltoro glacier

The ASTER imagery covered the lowermost 13 km of Baltoro glacier and associated tributaries. These areas tend to be heavily covered in debris, and their distinctive surface patterning makes them ideal for image matching. The velocity directions show fine details as the tributary glaciers round corners to join the main Baltoro glacier, and as the ice is channeled by the surrounding topography (Fig. 2). There is a general decrease in velocity towards the margins and termini of the glaciers, with the highest velocities reaching >200ma^–1 on icefalls in the upper parts of Uli Biaho and

Trango glaciers. Velocities on the main Baltoro glacier average 50 ma^–1, ranging from ∽75 ma^–1 in the upper part of the study area to <15 ma^–1 within the lowermost 2–3 km of the glacier.

As a check on the accuracy on the feature-tracking results, comparisons are made with the GPS-derived velocities from summer 2005 (Table 1; Fig. 2). In general, there is a close correspondence between the velocities measured at the six GPS locations and the feature-tracking results, particularly in relation to direction, with <5% variation in measured direction between the two methods. There is greater variability in the magnitudes of the velocities, these being on average higher in the GPS measurements than in the feature-tracking results. However, most of the recorded differences are within the estimated feature-tracking error limits of ±15ma^–1. The only exception is point 6, which was significantly faster in the GPS measurements. Reference MasonMayer and others (2006) also recorded GPS (summer) velocities that were greater than the annual average derived from manual feature tracking on Baltoro glacier; summer motion was up to 100% higher than winter motion in the upper parts of the ablation area (above the region shown in Fig. 2). We did not see such dramatic summer increases in velocity in this study, however, probably because the position of the highest-elevation GPS points in this study (5 and 6) is where the lowest-elevation GPS points started in the study undertaken by Reference MasonMayer and others (2006).

These results indicate that Baltoro glacier experiences spatial and temporal variations in ice motion, but that these variations are not necessarily synchronous across the glacier. Higher summer velocities are commonly observed on temperate and polythermal glaciers, and are usually attributed to basal lubrication caused by subglacial water flow (Paterson, 1994). This seems likely at Baltoro glacier, as there are many moulins and crevasses across the ablation area where meltwater can reach the glacier interior and bed, and surface melting is widespread in the summer. In summer 2005, for example, surface melt rates of 6.5 cmd^–1 were recorded close to GPS point 4, and 5.9 cm d^–1 2 km upstream of GPS point 6. The cross-profile B–B′ shows a rapid increase in velocity away from both margins, which suggests that basal sliding is the dominant motion mechanism in this part of the glacier (Fig. 2). Profile A–A′ shows quite a different pattern, however, with a more parabolic shape as velocities gradually rise to a peak in the center of the glacier (Fig. 2). This suggests that ice deformation is a more dominant motion mechanism close to the terminus, although basal sliding could still be important in the center of the glacier where subglacial water flow would presumably concentrate and increase basal lubrication. The very low velocities close to the margin at location A probably represent the influence of Trango glacier entering the main valley and slowing the ice immediately upstream. The greater importance of internal deformation towards the glacier terminus ties in with the GPS observations (described above) that significant increases in summer motion only occur in the upper part of our study area, and in the regions above it recorded by Reference MasonMayer and others (2006).

Overall, the GPS observations indicate that the feature-tracking results provide a realistic estimate of annual velocity patterns, both in terms of direction and magnitude. This gives confidence in the application of the technique to the other glaciers in this region.

4.2. South Skamri glacier

South Skamri glacier is located ∽30 km north of the Baltoro glacier drainage basin. It is a tributary of Skamri (Yengisogat) glacier (Fig. 3a). Historical Landsat imagery indicates a complex flow history between these glaciers. In 1978, the distortion of surface moraines suggests that Skamri glacier was surging, or had recently surged, and that it was the dominant flow unit in the basin (Fig. 3b). At this time, it pushed aside most of South Skamri glacier, effectively cutting it off from its lower terminus.

Fig. 3. (a)Velocity patterns across South Skamri and Skamri glaciers derived from feature tracking (current boundary between glaciers marked with red line). Inset shows velocity long-profile marked by white line. Arrows indicate location of feature-tracking match points and calculated flow direction. Velocities not shown where <15ma^–1 or where feature tracking was not successful. (b) Landsat image (18 July 1978) of the same region. (c) ASTER image (27 June 2007) used in (a), but without superimposed velocities.

By 2007, it had become clear that South Skamri glacier has surged, pushing Skamri glacier aside and now making South Skamri the dominant ice-flow unit in the basin (Fig. 3c). The feature-tracking results indicate that South Skamri glacier was active throughout its ablation area at this time, with velocities >250ma^–1 over an icefall in the upper part of its ablation area (Fig. 3a). The entire lower glacier is heavily debris-covered, with the velocities gradually decreasing in an along-glacier direction and towards the margins. In general, there are high and constant velocities over most of the upper ablation area where the ice flow is channelized, with a rapid reduction in flow (from >200ma^–1 to <100ma^–1 over a horizontal distance of ∽1 km) as the ice exits the main channel and spreads out over the wider lower valley (Fig. 3a inset).

An interesting feature of the velocity patterns shown in Figure 3a is that the boundary between South Skamri and Skamri glaciers occurs several kilometers upstream of the current end of the actively flowing ice (>15ma^–1). The velocity magnitudes and directions suggest that ice input from South Skamri glacier is now causing Skamri glacier to move downstream of where the glaciers join, as there is little evidence for current inputs from the main Skamri valley. When comparison is made with earlier imagery, it is clear that Skamri glacier provided the main driver of ice flow over much of the terminus prior to 1978 (Fig. 3b), but that the main flow driver has now switched to South Skamri glacier. Beyond the currently active ice shown in Figure 3a is a large area of stagnant ice at the glacier snout. This was derived originally from South Skamri glacier, but was effectively cut off by the 1978 (or earlier) surge of Skamri glacier.

In terms of the dominant ice-motion mechanism, the rapid increase in velocities away from the glacier margins and towards the glacier center in the upper part of South Skamri glacier is suggestive of basal sliding in this area. By contrast, the gradual reduction in velocities across the lower ablation area is more suggestive of deformation flow. This ties in with the large, dead ice areas in front of the existing active ice front, which are currently receiving insufficient mass to drive ice flow in that area.

4.3. Choktoi/Panmah glacier

Choktoi glacier forms the upper part of Panmah glacier, and their combined ablation area extends ∽20 km (Fig. 4a). The ice velocities derived from feature tracking along Choktoi glacier average ∽100ma^–1 over the central part of the ablation area, with a gradual decrease to zero over the lowermost ∽8 km where it forms Panmah glacier (Fig. 4a inset). Velocity estimates for the upper ∽5 km of the D–D′ profile shown in Figure 4a have a high degree of uncertainty due to the limited number of match points over this region, with the dotted purple line providing a best estimate of velocities over this area. The overall average velocity for the 18km profile shown in Figure 4a is 80 ma^–1, which suggests an average residence time of ∽225 years for the ice passing through the ablation area of the Choktoi/Panmah basin. Ice velocities increase rapidly away from the margins across the central part of Choktoi glacier, which suggests that basal sliding (blockschollen) is the dominant motion mechanism for this basin.

Fig. 4. (a)Velocity patterns across the Panmah glacier region derived from feature tracking. Arrows indicate calculated flow direction and are spaced every 25 pixels. Inset shows velocity long-profile marked by D–D^′ with dotted purple line indicating average velocity over upper ∽5 km where fewer feature-tracking match points were found. Velocities not shown where <15ma^–1 or where feature tracking was not successful. (b, c) Advance of First Feriole glacier, 2006–07. (d, e) Advance of Drenmang glacier, 2006–07.

Nobande Sobonde glacier, to the north of Panmah glacier, is largely inactive over the area shown in Figure 4a, except where tributaries join the main trunk. As discussed by Reference HewittHewitt (2007), all of these tributaries have surged in the recent past. Chiring glacier surged in 1995, whereas the adjacent South Chiring (Maedan) glacier surged sometime between 2002 and 2005. South Chiring glacier still eclipses the flow from Chiring glacier, with measured velocities of up to 60ma^–1 over its terminus region. However, velocities could not be derived for the upper ablation areas of Chiring, South Chiring and Second Feriole glaciers due to the low surface debris cover and few crevasses in these regions that precluded the finding of feature-tracking match points.

Second Feriole (Shingchukpi) glacier started surging in the fall or winter of 2004/05, and the velocity vectors in Figure 4a indicate that this surge pushed the ice of Nobande Sobonde glacier to the side (as also discussed by Reference HewittHewitt, 2007). The measured velocities on Second Feriole glacier are very similar to those on South Chiring glacier, suggesting that the surging had probably stopped by 2006–07.

Unlike the relatively low velocities measured for the tributaries discussed above, the Drenmang and First Feriole glaciers showed very high local velocities of >200ma^–1 at their termini over 2006/07. Reference HewittHewitt (2007) states that Drenmang glacier probably started surging in fall or winter 2004/05, with the glacier in summer 2005 overriding lateral moraines that had been ice-free for decades. Prior to this, Drenmang glacier last surged in 1977–78, with imagery from 1993 suggesting that the terminus moved at an average rate of ∽500ma^–1 in the 15 years after this event (Reference HewittHewitt, 2007). From the velocity patterns shown in Figure 4a it is clear that the current terminus is very active, with Figure 4d and e showing the advance of Drenmang glacier between summer 2006 and 2007. Distortion and folding of the medial moraines is also evident in these images, which provide further evidence of recent surging. Reference HewittHewitt (2007) argues that the 2004–05 surge probably started in the eastern branch, and today it is this tributary that dominates flow out of the basin, with the terminus pushing into the main trunk of Nobande Sobonde glacier and constructing its flow (Fig. 4d and e). One important distinction to make is whether the 2006–07 feature-tracking velocities are indicative of a continuing surge of Drenmang glacier that started in 2004–05, or whether they represent a post-surge relaxation phase of enhanced velocities. The latter explanation seems more likely, as actual Karakoram surges typically last for only a few weeks to months (e.g. 3 months for a 12 km surge of Kutiah glacier (Reference BurbankDesio, 1954); 8 days for a 3.2 km surge of Yengutz glacier; and 2.5 months for a 9.7km surge of Hassanabad glacier (Reference Goudie, Jones, Brunsden and MillerHayden, 1907)). Furthermore, Reference HewittHewitt (2007) states that it can take glaciers many years to return to pre-surge conditions after a surge has occurred. Moreover, the terminus velocities of ∽250ma^–1 are in the range likely for a post-surge slowdown, rather than full surge conditions.

Although a surge has not previously been recorded for First Feriole glacier, from visual inspection of the ASTER scenes (Fig. 4b and c), and from the high near-terminus velocities recorded by the feature tracking (Fig. 4a), it is clear that the glacier is currently very active. The terminus advanced ∽250m over the 2006–07 measurement period, was very steep and bulbous, and the ablation area appears to have thickened during this time (Fig. 4b and c). As discussed above, the three glaciers immediately to the north of First Feriole glacier all surged in the previous decade or so (Reference HewittHewitt, 2007), and old Landsat imagery (not shown) indicates that the glacier terminus was ∽3 km advanced from its present position in 1978 when surrounding glaciers were all retreating. These disparate sources of evidence all point to the fact that First Feriole glacier is likely to be a surge-type glacier, although we cannot discern whether the full surge process, or just a portion of it, was captured by the 2006–07 imagery.