Evolution of surface velocities and ice discharge of Larsen B outlet glaciers from 1995 to 2013

. We use repeat-pass SAR data to produce detailed maps of surface motion covering the glaciers draining into the former Larsen B Ice Shelf, Antarctic Peninsula, for different epochs between 1995 and 2013. We combine the velocity maps with estimates of ice thickness to analyze ﬂuctuations of ice discharge. The collapse of the central and northern sections of the ice shelf in 2002 led to a near-immediate acceleration of tributary glaciers as well as of the remnant ice shelf in Scar Inlet. Velocities of most of the glaciers discharging directly into the ocean remain to date well above the velocities of the pre-collapse period. The response of individual glaciers differs and velocities show signiﬁcant temporal ﬂuctuations, implying major variations in ice discharge as well. Due to reduced velocity and ice thickness the ice discharge of Crane Glacier decreased from 5.02 Gt a − 1 in 2007 to 1.72 Gt a − 1 in 2013, whereas Hektoria and


Introduction
Atmospheric warming and changes in ocean conditions during the past decades led to wide-spread retreat of ice shelves around the Antarctic Peninsula (API) (Cook and Vaughan, 2010). Progressive retreat culminated in the final disintegration of the Larsen A Ice Shelf in January 1995 and of the northern and central sections of the Larsen B Ice Shelf in March 2002 (Rott et al., 1996;Rack and Rott, 2004;Glasser and Scambos, 2008). The glaciers flowing from the Antarctic Peninsula plateau, previously feeding the ice shelves, became tidewater glaciers. Most of these glaciers accelerated significantly, resulting in increased ice discharge (Rott et al., 2002;De Angelis and Skvarca, 2003;Rignot et al., 2004;Scambos et al., 2004). The response of these glaciers to ice-shelf disintegration is of particular interest not only for quantifying the contributions of API outlet glaciers to sea level rise, but also for studying processes of ice shelf retreat and its interactions with grounded ice (Vieli and Payne, 2005;Hulbe et al., 2008).
Investigations on retreat and acceleration of glaciers in the Larsen Ice Shelf region so far focused mainly on the Larsen B embayment. Rignot et al. (2004) and Scambos et al. (2004) reported on acceleration of main glaciers draining into the Larsen B embayment, based on analysis of satellite images. Rott et al. (2011) derived velocities of nine Larsen B glaciers in pre-collapse state and in 2008 and 2009 from high-resolution radar images, and estimated calving fluxes and mass balance. Estimates of the mass balance of Larsen B glaciers in recent years have been derived from changes in surface topography. Shuman et al. (2011) and  tracked elevation changes over the period 2001 to 2009 using optical stereo imagery and laser altimetry of ICE-Sat (Ice, Cloud and Land Elevation Satellite) and of the ATM (airborne thematic mapper) sensor. Shuman et al. (2011) reported a combined mass loss of 8.4 Gt a −1 for these glaciers for the period 2001 to 2006, excluding ice lost by frontal retreat. Berthier et al. (2012) explained that the mass loss of former Larsen B tributary glaciers continued at almost the same rate over the period 2002 to 2011, reporting a mass loss rate of 9.04 Gt a −1 for the period 2006 to 2010/2011. Scambos et al. (2014) used satellite laser altimetry and satellite stereo-imagery to map ice elevation change and inferred mass changes for 33 glacier basins of the northern API over the time span 2001-2010. They report a mass balance of −7.9 Gt a −1 for the tributaries to the Larsen B embayment and −1.4 Gt a −1 for the tributaries to the remnant ice shelf in Scar Inlet.
These reports provide estimates of mass depletion for the Larsen B tributaries integrated over multi-year periods. Here we present a new analysis of satellite data showing the spatial and temporal variability in velocities over the whole Larsen B area dating back to 1995. We have included new satellite data not used in any previous studies so far, and have also reprocessed satellite radar images to generate fully consistent and comparable data sets on surface velocities. Our work includes both recent acquisitions by high resolution radar sensors as well as archived data, some of which have not been exploited until now. Velocity data and estimates of ice thickness are used to derive ice discharge at different epochs, showing significant temporal variability as well. The data sets provide a comprehensive basis for studying the dynamic response of the ice masses to the disintegration of Larsen B, including the glaciers that are draining now directly into the ocean as well as the remnant ice shelf in Scar Inlet and its tributary glaciers.

Data and methods
We derived maps of ice flow velocities from repeat-pass Synthetic Aperture Radar (SAR) data of the satellite missions ERS-1, ERS-2, Envisat, TerraSAR-X (TSX), and ALOS, applying either offset tracking or SAR interferometry (InSAR). The source data were obtained from the archives at the European Space Agency (ESA) and the German Aerospace Center (DLR). We retrieved two-dimensional surface displacement in radar geometry which we projected onto the surface, defined by the ASTER-based Antarctic Peninsula digital elevation model (API-DEM) of Cook et al. (2012), in order to produce maps of surface velocities. The maps of the surface velocity vector are provided in Antarctic polar stereographic projection resampled to a 50 m grid. The DEM is compiled from ASTER scenes from a range of dates between 2000 and 2009 which are unspecified in the final product (Cook et al., 2012). During this period various glaciers have been subject to major drawdown. The sensitivity analysis on the impact of possible DEM errors shows that even in extreme cases of surface lowering the induced error in geocoded velocity is below 1 %.
The spatial resolution of the SAR images along the flight track and in radar line of sight (LOS) ranges from 3.3 m × 1.2 m for TSX to 5.6 m × 9.6 m for the Advanced Synthetic Aperture Radar (ASAR) of Envisat. The time span of the repeat pass image pairs ranges from one day for ERS-1/ERS-2 tandem images to 46 days for ALOS Phased Array L-band SAR (PALSAR) images. Because of temporal decorrelation of the phase of the backscatter signal the interferometric (InSAR) method could only be applied for ERS-1/ERS-2 tandem images, available on several dates of the years 1995 to 1999. InSAR data of a single swath provide the surface displacement in LOS. We combined image pairs of ascending and descending orbits to derive 2-D velocity fields for the period late 1995 to early 1996. Being well before the collapse of the Larsen B Ice Shelf, this period is of particular importance as reference for studying the impact of ice shelf disintegration on tributary glaciers. For 1999, ERS SAR data were available only from single view direction. Assuming unaltered flow direction since 1995/1996, we derived velocity maps in November 1999, using the argument from the velocity vectors of crossing orbits.
For retrieving maps of ice motion from the TSX SAR, Envisat ASAR and ALOS PALSAR we apply the offset tracking technique which is based on cross-correlation of templates in SAR amplitude images. Offset tracking delivers along track and LOS velocity components from a single image pair. It is less sensitive to displacement than InSAR, but this drawback is (at least partly) compensated by the longer time span between the repeat pass images (Rott, 2009). We used templates of 64 × 64 and 96 × 96 pixels size and applied sampling steps of 10 pixels for generating velocity maps. TSX images are our main data sources for velocity maps between 2007 and 2013, complemented by occasionally available ALOS PALSAR data. Envisat ASAR data are the basis for velocity maps for 2003 to 2006 on large glaciers and on the Scar Inlet ice shelf.
The uncertainty of retrieved velocities differs between the sensors. The ERS InSAR motion maps are based on InSAR pairs of good coherence. One fringe (phase cycle of 2 π ) corresponds to 7.2 cm projected onto a horizontal surface. Assuming an uncertainty of 0.2 fringes for a point on the moving glacier surface and 0.2 for the zero velocity reference points on ice free surfaces, for ERS InSAR the uncertainty in surface velocity of grounded ice is ±0.02 m d −1 . On floating ice control points without horizontal motion are used as reference, so that the observed signal corresponds to the tidal displacement. The phase differences between individual reference points, located around the Seal Nunataks and in inlets along Jason Peninsula, are less than 0.5 fringes. Assuming an uncertainty of 0.2 fringes for the moving ice shelf and of 0.5 fringes for the reference points, the uncertainty in horizontal velocity of floating ice is ±0.04 m d −1 .
For offset tracking, the accuracy depends on the pixel size, the time interval, and the quality of features in order to obtain good correlation peaks. We excluded areas of low correlation, so that the uncertainty for the retrieval of displacement is in the order of 0.2 to 0.3 pixels. The resulting uncertainties in the magnitude of surface motion are ±0.05 m d −1 for TSX SAR, ±0.08 m d −1 for ALOS PALSAR and ±0.15 m d −1 for Envisat ASAR.
The mass flux across a gate of width Y [m] near the calving front or grounding line is computed according to where ρ ice is the density of ice, u m is the vertically averaged horizontal velocity, θ is the angle between the velocity vector and the gate, H is the ice thickness. We use a columnaveraged ice density of 900 kg m −3 to convert ice volume into mass. For calving glaciers, full sliding is assumed across calving fronts, so that u m corresponds to the surface velocity, u s , obtained from satellite data. For glaciers discharging into the ice shelf, we estimate the ice deformation at the flux gates applying the laminar flow approximation (Paterson, 1994) using a rate factor as derived by Hulbe et al. (2008) for outlet glaciers to Larsen B. The results show moderate values of deformation velocities. For Crane Glacier, the resulting vertically averaged velocity (pre-collapse) is u m = 0.96 u s ; for other glaciers, u m = 0.95 u s . Ice thickness at the flux gates is obtained from various sources. For Flask and Starbuck glaciers, radar sounding data are available (Farinotti et al., 2013. For Crane Glacier, the cross section of the calving gate is deduced from bathymetric data (Zgur et al., 2007;Rott et al., 2011). For Leppard Glacier, ice thickness data of Huss and Farinotti (2014) are used. For calving fluxes of Crane, Hektoria and Green glaciers, the ice thickness in the center of the flux gate is estimated from surface height above sea level assuming flotation. The central sections of these glacier fronts have been floating at least since 2007. The surface elevation near the calving front is obtained from laser ranging data of ICESat and the ATM Thomas, 2013, 2014) and in 2011 and 2013 also from digital elevation data of TanDEM-X (Krieger et al., 2013). For uncertainty estimates of mass fluxes through the gates, we assume ±10 % error of the cross section area for Starbuck, Flask and Crane glaciers, and ±20 % for Hektoria, Green, Jorum and Leppard glaciers. For velocity across the gate, we assume ±5 % uncertainty.

Velocities and frontal retreat of glaciers draining into Larsen B embayment
The location of the glacier basins is shown in Fig. 1, and the areas of the basins for the region upstream of the 1995 grounding line and of the 2012 glacier fronts are specified in Table 1. The basin outlines inland were provided by A. Cook based on the ASTER derived Antarctic Peninsula DEM (API-DEM) (Cook et al., 2012). The positions of the grounding lines in 1995 are from the ERS InSAR analysis of Rack (2000). The update of glacier fronts and areas in 2012 is based on a Landsat image of 12 January 2012. Before 2002 all glaciers between the Seal Nunataks and Jason Peninsula drained into Larsen B Ice Shelf. Since its collapse, in March 2002, they drain into a wide bay and in the remnant part of the ice shelf in Scar Inlet. The area of the Larsen B tributary glaciers decreased by 270 km 2 since 1995. The 2012 area refers to the ice front rather than the grounding line, so that the total loss in grounded ice extent is slightly higher because frontal sections of some glaciers are floating.
The largest glaciers north of Scar Inlet, where the ice shelf disappeared in 2002, are the Hektoria, Green, Evans (HGE) and Crane glaciers. Before the ice shelf breakup the frontal zone of HGE glaciers was formed by the confluence of the three glaciers, stretching across a wide bay. Following the ice Table 1. Area of glacier basins (in km 2 ) shown in Fig. 1 above the October 1995 grounding line and updated for glacier fronts on 12 January 2012, and change of glacier area 1995 to 2012. NC -no significant change of front position or grounding line. The areas of glacier basins include rock outcrops and mountain slopes.

Nr. Glacier
Area ( shelf collapse the frontal regions of HGE retreated quickly (Rack and Rott, 2004;Scambos et al., 2004), suggesting that they were lightly grounded and sensitive to changes in iceshelf buttressing. The ice shelf collapse resulted in the progressive breakup of increasingly large areas of grounded ice concomitant with acceleration of ice flow and dynamic thinning, amounting to a total retreat of 174 km 2 by January 2012. On Crane Glacier, the loss of grounded ice has been smaller (35 km 2 ) because the terminus is confined in a narrow fjord. Jorum Glacier lost 24 km 2 in grounded ice; Punchbowl Glacier lost 12 km 2 , and Melville Glacier lost 4.1 km 2 . The frontal positions of Mapple and Pequod glaciers have been stationary. An overview map of surface velocities for the Larsen B region is shown in Fig. 2a for the year 1995 based on ERS InSAR data. As already reported by Rott et al. (2011), the 1995 velocities of outlet glaciers to Larsen-B agree within a few percent with the velocities retrieved from 1999 InSAR data. There is no indication for a significant temporal trend in velocity on any of the glaciers. The velocities, derived from InSAR data on various dates in 1995 and 1999 differ by less than 5 % at any of the flux gates. Varying tidal deformation along the ice shelf margins, observed in the different interferograms, did not affect the ice motion at these flux gates which are located several kilometers inland of the 1995-1999 grounding zone. Figure 2b is a composite of several velocity maps from TSX and ALOS PALSAR offset tracking analysis of the years 2008 to 2012. As the figures show, a major flow acceleration is observed for HGE, Jorum, and Crane glaciers. Flask and Leppard glaciers in Scar Inlet also accelerated, but at a lower rate. In order to investigate the temporal evolution of velocities we extracted profiles along the central flow line of the main glaciers: Hektoria, Green, Jorum, Crane, Punchbowl and Melville glaciers, now terminating with calving fronts (Fig. 3), and Flask and Leppard which are still confined by the remnant part of Larsen B Ice Shelf. The location of the profiles is charted in Fig. 1. The map of velocity changes (Fig. 2c) and the longitudinal profiles show that the flow acceleration extends far upstream on the large glaciers, whereas on the smaller glaciers the acceleration has been modest and confined to the lower part of the tongues.
The velocity of Hektoria and Green glaciers is presently still much higher than in 1995, but has been subject to strong variations since 2002 associated with glacier thinning and frontal retreat. The velocity profiles (Fig. 3) show periods of acceleration followed by gradual deceleration. In 2008 Hektoria and Green glaciers still had a common terminal section, but the lower terminus was already heavily fractured (Fig. 4).  . In spite of continued thinning, although with reduced rate, the position of the glacier front has been rather stable since 2006. The shape of the glacier bedrock in form of a deep canyon, inferred from bathymetric data, indicates that the central part of the lower terminus has been ungrounded for several years (Rott et al., 2011). This suggests that lateral drag plays a key role in maintaining the frontal position since 2006. Also the velocity of the Jorum Glacier terminus is still higher than before ice shelf collapse. The velocity variations of the outlet glaciers are clearly dominated by multi-annual trends triggered by ice shelf dis-integration. On some of the glaciers, seasonal variations in velocity by a few per cent are observed, but not in every year. Compared to the long term trend this signal is not significant.

Velocities of Scar Inlet ice shelf and tributary glaciers
The  (Fig. 6), another indication that the Larsen B disintegration event had a rather immediate impact on the stress field of Scar Inlet ice shelf.
In the profiles of 1995 and 1999, based on one-day InSAR repeat pass data, we exclude the tidal deformation zone because of ambiguity between horizontal motion and vertical displacement.
In spite of still being backed up by an ice shelf, both Flask and Leppard glaciers accelerated since 1995/1999 (Fig. 6). Between 1995 and 1999, there are no apparent differences in velocity. On Flask Glacier, the mean velocity in 2009-2013 at the flux gate, 6 km above the grounding line, is 41 % higher than the velocity in 1995/1999. On Leppard Glacier,

Temporal variations of ice discharge
Estimates of ice discharge of Crane, Jorum, Hektoria, and Green glaciers in different years are presented in Table 2. The estimated discharge of Hektoria and Green glaciers for 1995 amounts to 1.19 Gta −1 using the same gate near the 2008 front as Rott et al. (2011) (Fig. 4). By February/March 2004, 2 years after the collapse, the maximum velocity at this gate was 5.1 m d −1 (1862 m a −1 ), 5 times higher than in 1995. A transect on Hektoria Glacier, acquired by the NASA ATM in 2004 (Krabill and Thomas, 2013), allows for an estimate of an ice thickness of 406 m under the assumption of flotation, resulting in a flux of 4.74 Gt a −1 . The estimate for 2008 by Rott et al. (2011) amounts to 2.88 Gt a −1 . At that  (Fig. 4). A transect of surface elevation on Hektoria Glacier was measured in 2011 by the ATM during the Ice-Bridge campaign. The freeboard at the gate is 55 m, resulting in a maximum ice thickness of 450 m assuming freely floating ice. The corresponding numbers for the calving fluxes, with November 2010 velocities, are 1.67 Gt a −1 for Hektoria Glacier and 1.99 Gt a −1 for Green Glacier, adding up to 3.66 Gt a −1 which is 27 % higher than the flux in 2008. By July 2013 the combined flux decreased to 3.05 Gt a −1 . This illustrates the impact of velocity variations on calving fluxes, resulting in major fluctuations of glacier net mass balance within a few years.
For computing the ice flux for Crane Glacier, the same flux gate 1 km inland of the ice front in 2008 and 2009 is used as by Rott et al. (2011) (Fig. 4). Because of slow down of ice flow (Fig. 3) and reduction in ice thickness, the calving flux of Crane Glacier decreased significantly during recent years. Based on the June 2007 analysis, the flux across the gate is estimated at 5.02 Gt a −1 , 4.4 times higher than the pre collapse calving flux of 1.15 Gt a −1 . Until November 2013 it decreases to 1.72 Gt a −1 , one third of the 2007 flux.
In 1995 the combined mass flux of Jorum Glacier across the 2008 calving gates of the two glacier branches amounted to 0.35 Gt a −1 (Rott et al., 2011). For 2008, the elevation data from an ICESat transect close to the gates were used to estimate the maximum ice thickness. For estimating the ice thickness in 2012, we use surface elevation data of the TanDEM-X satellite mission, which show surface lowering by a few meters since 2008. The estimated calving For Starbuck, Flask and Leppard glaciers, data on ice thickness are available from ice sounding measurements and ice flow modeling (Farinotti et al., 2013Huss and Farinotti, 2014). On Starbuck Glacier, the TSX ice motion data of 2009 and 2011 do not reveal any significant difference compared to 1995 (Fig. 7). Therefore the discharge has likely not changed significantly either. The flux through the cross section near the grounding line, with maximum velocity of 0.34 m d −1 (124 m a −1 ), is estimated at 0.067 Gt a −1 ( Table 3). The ice on Flask and Leppard glaciers is thicker and velocities are higher. For Flask Glacier, the mass flux is derived for a gate along a transverse profile 4 km above the grounding line (Fig. 7). This corresponds to the position of radio echo sounding profile 1, acquired by the BAS Polarimetric Airborne Survey Instrument in November 2011 (Farinotti et al., 2013). In the center of the profile the ice thickness is 690 m. In 1995 and 1999, the velocities in the center are 1.31 (478 m a −1 ) and 1.36 m d −1 (496 m a −1 ), respectively, the resulting ice discharge across the gates is 0.78 Gt a −1 and 0.80 Gt a −1 for the 2 years. On Flask Glacier, the velocities in 2009-2013 vary between 1.76 (642 m a −1 ) and 1.93 m d −1 (704 m a −1 ), and the ice discharge ranges from 1.0-1.23 Gt a −1 , without a clear temporal

Discussion
In line with previous studies, our data shows a drastic increase in flow velocities of major tributary glaciers following the collapse of Larsen B Ice Shelf in early 2002. Reduced back stress and frontal retreat caused flow acceleration that propagated up-glacier. Beyond that, our analysis of new velocity data shows that some of the glaciers slowed down significantly during recent years. Strong acceleration and increase of calving flux is observed for HGE glaciers and Crane Glacier, downstream of which the seafloor map shows deep troughs (Lavoie et al., 2015). Scambos et al. (2004)    Subsequently, our analysis shows significant deceleration for Crane Glacier since mid-2007, yet over this time period the position of the ice front has remained comparatively stable. Possibly this is due to a reduction in the ratio between driving stress and lateral shear, in accordance with decreasing surface slope on the lower glacier terminus. Targeted ice-flow modeling is required to further address this issue. From June 2007 to November 2013 the calving flux of Crane Glacier decreased from 5.02 to 1.72 Gt a −1 . Under the assumption that the pre-collapse flux corresponds to the bal- ance flux (Rott et al., 2011), the resulting rate of mass loss decreased from 3.87 to 0.57 Gt a −1 . Based on differencing of DEMs from optical stereo imagery in combination with ICESat data, Scambos et al. (2014) report a mean loss rate of 2.24 Gt a −1 for the period March 2003-November 2008. This is 42 % lower than our estimate for June 2007 and 27 % higher that our estimate for 2008/09. These large temporal variations emphasize the importance of using common epochs when comparing glacier contributions to sea level rise obtained by different methods.
Whereas on Crane Glacier a period of major flow acceleration during the first 5 years after ice shelf disintegration was followed by a steady gradual decrease in velocity, the flow behavior of Hektoria and Green glaciers has been more variable. Periods with increased flow velocities and frontal retreat alternated with periods of comparatively stable front positions or short-term advance. ASTER and ICESat data show substantial elevation losses on lower Green Glacier amounting to about 100 m during the time span November 2001-late 2008 . Scambos et al. (2014) Table 3 referring to gates shifted inland because of frontal retreat.
Hektoria, Green and Evans glaciers, forming a joined terminus in a wide bay in 2002, have been particularly vulnerable to stress perturbation after ice shelf collapse as evident from the frontal retreat. Successive phases of transition from weakly grounded to floating ice due to flow acceleration and thinning, associated with major calving events, have been maintaining high rates of mass depletion for the HGE glaciers to date. Crane and Jorum glaciers, terminating in deep and narrow fjords, have been subject to acceleration and major mass depletion during the first 5 years after ice shelf collapse, but slowed down afterwards. Similar behavior after retreat into narrow fjords is observed for Sjögren and Boydell glaciers in the Prince Gustav Channel and for the Dinsmoor, Bombardier and Edgeworth glaciers in the Larsen A embayment (Rott et al., 2014). The ratio of longitudinal stress to lateral shear stress is critical for glacier motion in narrow valleys (Hulbe et al., 2008). Decreasing ice thickness and surface slope affect driving stresses and cause deceleration in flow. However, considering the ongoing thinning of the terminus and the resulting decrease of lateral shear stress, it can be concluded that Crane and Jorum glaciers will still be subject to major retreat before reaching a new equilibrium state further inland.
The two main glaciers draining into Scar Inlet ice shelf, Flask and Leppard glaciers, have also been affected by flow acceleration in recent years. GPS measurements at stakes on Larsen B located 50 km downstream of these glaciers showed flow acceleration in the order of 10 % between 1994 and 1999 (Rack, 2000). This indicates that also the southern sections of Larsen B Ice Shelf had weakened mechanically previous to the disintegration event in 2002, as reported for the northern and central sections Rack and Rott, 2004). Our analysis of substantial flow acceleration and development of rifts, evident in satellite data of 2004, implies that the break-up had a near immediate impact on the stress field of the ice shelf. Fricker and Padman (2012) report for two crossover points on Scar Inlet ice shelf relatively constant elevation change of ∼ −0.19 m a −1 during 1992-2008. Our analysis of the temporal evolution of ice shelf flow suggests that changes in the rheology and stress field might not have been continuous during this period. The main speedup on Scar Inlet ice shelf occurred during the first 2 years after the disintegration of the northern and central sections of Larsen B, whereas changes later on were more gradual. Given the spatial pattern of acceleration, with main speedup in the ice shelf section nourished by Flask and Leppard glaciers, further weakening has to be expected along the shear margins of this section, as well as for the ice imme-diately downstream of the grounding zone. Numerical models of Larsen B Ice Shelf in pre-collapse state show a band of weak ice along the shear zone that separates the outflow of Leppard Glacier from the slowly moving ice along Jason Peninsula Vieli et al., 2006). The differential acceleration of flow and the formation of additional rifts, which are evident in ASAR and TSX images since 2004, indicate that ice in this zone is further weakening.
Whereas the ice at the flux gates of Leppard and Flask glaciers accelerated from 1995 to 2009 by 44 and 38 %, respectively, the velocity of Starbuck Glacier has been stable. This can on one hand be attributed to the bedrock topography, on the other hand to the rather modest mass turnover. The lower terminus of Starbuck Glacier is firmly grounded, with a broad sub-glacial ridge in the area of the grounding zone . Under the assumption of mass balance equilibrium, supported by the observed steady ice motion since 1995, a specific surface mass balance b n = 230 kg m −2 is inferred from the ice flux across the grounding line.
The stable velocity in 1995 and 1999 suggests that Flask Glacier has been close to equilibrium state in those years. Thus, assuming equilibrium condition, the 1995 mass flux of 0.78 Gt a −1 across the flux gate results in b n = 779 kg m −2 , 3.4 times higher than the specific mass balance on Starbuck Glacier. The large difference in b n can be explained by the strong west-east decrease of accumulation (Turner et al., 2002). Flask Glacier flows down from the main ice divide of the peninsula, whereas Starbuck Glacier originates on a small ice plateau 25 km to the east, separated from the main divide by the deep trough of Crane Glacier.
Flask and Leppard glaciers have responded to the changing stress conditions on the ice shelf in front by acceleration. The bedrock of Flask and Leppard glaciers ascends towards the grounding zone from depressions several kilometers upstream (Farinotti et al., 2013;Huss and Farinotti, 2014). The height of the glacier surface above the bedrock suggests that the glaciers are still firmly grounded above at the flux gates. Consequently, changes in the force balance of the grounding zone probably played a main role for initializing flow acceleration. Scambos et al. (2014) report rates of mass change of

Conclusions
The collapse of the main section of Larsen B Ice Shelf in March 2002 triggered a near immediate response of most tributary glaciers with increased velocities maintained to date. Acceleration of ice flow is also observed on the remnant part of the ice shelf in Scar Inlet and its main tributaries. The behavior of the individual glaciers varies, and velocities show significant fluctuations over time. Whereas, after an initial speed up, Crane and Jorum glaciers slowed down significantly since mid-2007, the Hektoria and Green glaciers continue to show widespread fluctuations in velocity and periods of major frontal retreat alternating with more stationary positions or short term frontal advance. These differences in the response are related to glacier geometry and bedrock features. Crane and Jorum glaciers retreated into deep and narrow fjords while Hektoria and Green glaciers still calve into a wide bay. Temporal fluctuations of flow velocity are a main factor for fluctuations in ice discharge, emphasizing the importance of common epochs for reconciling glacier mass balance estimates derived by different methods (Shepherd et al., 2012). Because of the combined effect of slow down and decrease in ice thickness, the ice discharge of Crane Glacier decreased by 66 % between 2007 and 2013 and of Jorum Glacier by 26 %. Both glaciers are expected to retreat further inland before reaching a new equilibrium in spite of slow-down, concluding from ongoing thinning and the increase of floating ice area. Hektoria and Green glaciers maintained variable but consistently high rates of mass depletion in recent years, as the calving front alternated between floating and weakly grounded phases.
The increase of flow velocity on Scar Inlet ice shelf and its larger tributaries started soon after the 2002 Larsen B collapse event, but changes have been discontinuous with most of the increase in the first years followed by comparatively small variations in velocity since 2009. On the smaller tributaries, changes have been modest or absent. The velocity on the ice shelf section downstream of Flask and Leppard glaciers, the largest tributaries, increased 2-3-fold since 1995/1999. The velocity at the flux gates of these glaciers increased until 2009 by 38 and 44 %, respectively, with minor fluctuations in velocity in later years. This suggests that the Scar Inlet ice shelf and its main tributary glaciers may have temporarily adjusted to the loss of the back stress from the main Larsen B Ice Shelf. However, considering the sustained high flow velocities and the enhanced formation and extension of cracks along the shear margins of the central ice shelf section, this state will not be long-lasting. These are clear signs for flow instability that will very likely lead to a complete disintegration of Scar Inlet ice shelf in the near future.