Four-decade record of pervasive grounding line retreat along the Bellingshausen margin of West Antarctica

Changes to the grounding line, where grounded ice starts to ﬂ oat, can be used as a remotely sensed measure of ice-sheet susceptibility to ocean-forced dynamic thinning. Constraining this susceptibility is vital for predicting Antarctica ’ s contribution to rising sea levels. We use Landsat imagery to monitor grounding line movement over four decades along the Bellingshausen margin of West Antarctica, an area little monitored despite potential for future ice losses. We show that ~65% of the grounding line retreated from 1990 to 2015, with pervasive and accelerating retreat in regions of fast ice ﬂ ow and/or thinning ice shelves. Venable Ice Shelf confounds expectations in that, despite extensive thinning, its grounding line has undergone negligible retreat. We present evidence that the ice shelf is currently pinned to a sub-ice topographic high which, if breached, could facilitate ice retreat into a signi ﬁ cant inland basin, analogous to nearby Pine Island Glacier.


Introduction
Satellite remote sensing of the Antarctic Ice Sheet over the last 25 years has revealed trends of ice loss that are especially pronounced around its coastline [Shepherd et al., 2010;McMillan et al., 2014]. Manifested by thinning of floating ice [e.g., Paolo et al., 2015], dynamic thinning of grounded ice [e.g., Pritchard et al., 2009], and rapid retreat of grounding lines [e.g., Rignot et al., 2014], these ice losses form one of the largest components of contemporary global sea level rise [Shepherd et al., 2012;Vaughan et al., 2013]. However, the processes by which they occur remain imperfectly understood. The spatiotemporal distribution of the observed ice losses strongly suggests a forcing that initiates at ice-sheet margins , and several recent studies have attributed the dynamic thinning of grounded ice to the thinning of ice shelves MacGregor et al., 2012;Pritchard et al., 2012]. This is, in turn, a possible consequence of the ingress of relatively warm circumpolar deep water (hereafter CDW) across the continental shelf to the sub-ice shelf cavity [Walker et al., 2007;Jacobs et al., 2011;Bingham et al., 2012]. These processes are hypothesized to be responsible for ice losses from the sectors of West Antarctica that drain to the Amundsen and Bellingshausen Seas [Rignot et al., 2008;Pritchard et al., 2012;Bingham et al., 2012] but, while the former has received considerable observational and modeling attention [Payne et al., 2004;Shepherd et al., 2004;Scott et al., 2009;Joughin et al., 2012;Favier et al., 2014;Joughin et al., 2014;Rignot et al., 2014;Seroussi et al., 2014;Goldberg et al., 2015], few observations of glacial change in the Bellingshausen Sea Sector (hereafter BSS) have been published.
It is important to generate improved records of glacial change in the BSS for several reasons. First, the estimated mass lost from the region between 1992 and 2006 (~49 Gt yr À1 ) comprised 35% of total mass loss from the West Antarctic Ice Sheet [Rignot et al., 2008], second only to contributions from the Amundsen Sea Sector (~64 Gt yr À1 ; 46% of total loss). Second, a recent study has suggested that mass loss since 2009 has been particularly significant and that the region may have destabilized [Wouters et al., 2015]. Third, although studies from the neighboring Amundsen Sea Sector are offering insights into dynamic thinning and its possible oceanic forcing, the scarcity of glacial change records and supporting information in the BSS (such as sub-shelf and grounded-ice geometry and physical oceanographic data) make it difficult to assess whether changes in the BSS have occurred at similar or different time scales to those in the Amundsen Sea Sector, potentially implicating different or additional forcing or feedbacks.
In order to improve our understanding of glacial change across this region, we generated a comprehensive data set of grounding line (hereafter GL) positional change along the Bellingshausen Sea coastline for the period 1975-2015. We exploited a combination of optical and radar satellite imagery to map GL positions at approximately 5 to 10 year intervals. We find that, over this 40 year period almost the entire length of • Supporting Information S1 • Table S1 • Table S3 • Data Set S1 the coastline has experienced GL retreat, with the greatest retreat typically occurring in regions containing deeply bedded, fast-flowing outlet glaciers and ice streams. We suggest that this pervasive trend implicates widespread access of CDW to the Bellingshausen glacial margin, so that even despite relatively high accumulation inland, the mass balance for much of the region is strongly negative. Only in some special cases has the GL remained relatively stable in position or undergone advance, probably as a result of local topographic pinning that is likely to be overcome with continued regional oceanic forcing.

Landsat Grounding Line Mapping
We used Landsat optical satellite imagery as our primary data source due to its unmatched spatial-temporal coverage compared with other remote sensing data sets covering the BSS. The position of the GL was mapped from these images at 5 to 10 year intervals between 1975 and 2015 as follows.
Referring to the conceptual diagram of Fricker et al. [2009, their Figure 2], the "true" GL (where ice decouples from the bed and begins to float due to tidal forcing; their "G") cannot be identified reliably with static, optical imagery (such as Landsat). Instead, using Landsat imagery, the most suitable proxy for G is the break-in-slope, otherwise known as the "inflection point" (hereafter I b ; after notation in Fricker et al. [2009]), which is defined as the most seaward continuous slope break detectable in satellite imagery [Brunt et al., 2010;following Scambos et al., 2007]. Situated within close proximity to G, I b appears as a clearly defined shadow-like change in image brightness on optical imagery [cf. Bindschadler et al., 2011] ( Figure S1 in the supporting information).
Using GIS tools on georeferenced multispectral Landsat scenes (Table S1), we digitized the position of I b at approximately 50-100 m intervals along the BSS coastline and coastal islands for years 1975, circa 1985, circa 1990, 2000, 2005, 2010, and 2015. The majority of scenes used were acquired during austral summer (January/February). For circa 1985 and circa 1990 we utilized data from ± 2 years where 1985/1990 scene spatial coverage was poor. Throughout the study, only images with cloud cover ≤10% were utilized. The launch failure of Landsat 6 in 1993 was responsible for a hiatus in Landsat acquisitions throughout the mid-1990s, precluding any mapping of results for circa 1995.
Using a similar I b mapping technique to that employed by Bindschadler et al.  Table S2]. In terms of the principal objective of this paper, which is to monitor changes in the position of the GL across 5 year intervals, any imprecision in locating I b (on the order of 10 2 m) is outweighed by changes in its position after 5 years (on the order of 10 2 -10 3 m). This uncertainty broadly matches the positional uncertainty associated with other remote sensing techniques, such as interferometric synthetic aperture radar (InSAR) [cf. .
It is important to recognize that under some conditions the use of I b as a proxy for the GL may fail [cf. Fricker et al., 2009]. The typical contexts in which I b makes a poor indicator are in areas of fast ice flow, where subglacial bed and surface slopes are shallow and I b is difficult to define; and/or in ice plains, where multiple breaks-in-slope may be present around the grounding zone, or where the break-in-slope may decouple substantially from the location of the true GL [cf. Corr et al., 2001;Fricker and Padman, 2006;Fricker et al., 2009;Brunt et al., 2011]. Therefore, to ascertain the utility of Landsat for monitoring GL change, we additionally employed InSAR mapping techniques to monitor the nature and configuration of grounding zones throughout the BSS and to identify the potential presence of ice plains in this sector.

Interferometric Synthetic Aperture Radar (InSAR)
Where synthetic aperture radar (SAR) observations have been acquired, it is possible to map the inland limit of tidal flexure acting on the grounding zone, F, directly [e.g., Fricker et al., 2009, Figure 2;Brunt et al., 2010;. We applied interferometry to SAR data acquired from the European Space Agency's ERS-1 and ERS-2 satellites to delineate F along the BSS coastline between 1992 and 2011. The data consist of SAR scenes acquired in 1992, 1994, 1996, and 2011 (Table S4), and the generated interferograms have temporal baselines of either 1 day (1996) or 3 days (1992, 1994, and 2011).

10.1002/2016GL068972
Following Park et al. [2013], we employed a double-differential InSAR processing technique, whereby we differenced consecutive interferograms, corrected for the effects of surface topography using a resampled version of the Bedmap2 surface digital elevation model [Fretwell et al., 2013], in order to remove signals related to ice flow and reveal vertical surface motion due to tidal flexure acting upon the grounding zone from floating ice shelves. Such motion, which is represented in double-differenced interferograms as a band of closely spaced fringes across the grounding zone ( Figure S2), reveals F as the landward limit of tidally induced vertical ice motion.
To quantify the positional uncertainty of InSAR-derived F, we compared the tidally variable location of F from multiple double-differenced interferograms, where possible across multiple epochs. From this, we estimated tidally induced variation in F to range from~± 100 m across areas of steeply bedded, shoaling subglacial topography to ≤~± 300 m across deeply grounded outlet glaciers with shallow bed slopes. These values align with positional accuracy estimations in other areas of Antarctica [cf. Park et al., 2013].
Our determination of InSAR-derived F confirms the absence of ice plains and other complex ice-shelf geometries along the BSS. Therefore, unlike for other regions of Antarctica where optical-based I b mapping campaigns have been prone to failure [e.g., Fricker et al., 2009;Brunt et al., 2011], we are confident that Landsat-derived I b acts as a good proxy for the GL in this sector.

Quantifying Grounding Line Position Retreat/Advance
To quantify GL advance or retreat over epochs, we defined the 2015 Landsat-mapped I b as a baseline. We partitioned this baseline into 30 km segments, and at the limit of each segment defined a normal extending infinitely both landward and seaward from the baseline. These normals intercept the mapped I b for all years.
To calculate advance or retreat of I b for any 30 km segment of coastline over any period [2015y], where y = year of interest, we defined a polygon bounded by the 2015 baseline, the mapped I b for year y, and the two intercepting normals. We then summed the areas of each polygon, in each case defining whether I b had retreated or advanced over the epoch of interest, and then divided by 30 km to convert the final figure into a magnitude of I b advance/retreat over each epoch of interest (see Text S2 and Data Set S1). Figure 1 shows that the most significant changes to the GL position between 1990 and 2015 are located at Ferrigno and Fox Ice Streams (À2.77 ± 0.50 km and À1.79 ± 0.14 km, respectively) and Stange Ice Shelf (À0.92 ± 0.14 km). Since 1990, net GL retreat has been widespread along the BSS coastline and around Thurston Island and has occurred along much of the margin draining to the Amundsen Sea between the Cosgrove Ice Shelf and Pine Island Glacier. In total, between 1990 and 2015, 65.4% of the GL along the mainland BSS experienced net retreat and only 7.4% net advance. Over the same epoch, 29.3% of the GL around Thurston Island experienced net retreat and only 1.8% net advance. Most of the detectable retreat around Thurston Island occurred along its southern (Abbot Ice Shelf-facing) margin; change along its northern margin was often not possible to discern outside Landsat 1σ-error bounds. Figure 1 also shows the tendency for the greatest GL retreat to be associated with regions of fast-flowing, likely deeply grounded ice.

Results
Partitioning the observations into change over 5 to 10 year epochs (Figure 2), the most comprehensive coverage along the entire coastline begins in 1990, from which time it is clear that the pace of GL retreat at Eltanin Bay has been persistently high. There (for segments 22-26 in Data Set S1; locations of Ferrigno and Fox Ice Streams), the GL retreated on average 0.45 ± 0.14 km (at a rate of 45 ± 27 m yr À1 ) from 1990 to 2000, 0.46 ± 0.14 km (92 ± 27 m yr À1 ) from 2000 to 2005, 0.50 ± 0.14 km (100 ± 27 m yr À1 ) from 2005 to 2010, and most recently 0.52 ± 0.14 km (104 ± 27 m yr À1 ) from 2010 to 2015.
Where data exist from 1975 or 1985 onward, some sites experienced relatively fast GL retreat in the earlier epochs we have analyzed, followed by slower retreat or undetectable change, and then increased retreat toward the present day. One example is the western Stange Ice Shelf, where the GL at Berg Ice Stream retreated 0.54 ± 0.14 km (36 ± 27 m yr À1 ) between 1975 and 1990, remained stable within Landsat error bounds between 1990 and 2010, then retreated 0.24 ± 0.14 km (48 ± 27 m yr À1 ) between 2010 and 2015.
Farther west, between latitudes 93°W (eastern Abbot Ice Shelf) and 87°W (Wiesnet Ice Stream/Venable Ice Shelf) we also find notable GL retreat between 1985 and 1990 relative to all intervening epochs until 2010-2015. In 2010-2015, the GL along eastern Abbot Ice Shelf then experienced a renewed retreat (distance 0.31 ± 0.14 km; rate 62 ± 27 m yr À1 ; segments 38-40 in Data Set S1), while the GL also experienced a detectable minor Geophysical Research Letters 10.1002/2016GL068972 retreat between the Wiesnet and Williams Ice Streams feeding Venable Ice Shelf (0.23 ± 0.14 km at 45 ± 27 m yr À1 ; segments 30 in Data Set S1).

Discussion
Our observations demonstrate that since 1975/1985 there has been a net retreat of the grounding line along almost all of the BSS (Figures 1, 2, and S3), demonstrating a pervasive trend of ice response along the whole Bellingshausen Sea coastline. Within this trend, several further phenomena are notable. First, both the greatest retreat (Figure 1) and the greatest increase in GL retreat rates from the 1990s to the present (Figure 2) have occurred along the eastern BSS coast, incorporating the glaciers and ice streams draining to Eltanin Bay and Stange Ice Shelf. Second, observed retreat rates in the BSS have varied over time across different locations. Third, while the ice streams feeding the Venable Ice Shelf have experienced limited retreat over the last three decades, retreat rates do not align with the more pronounced glaciological changes occurring more widely across the BSS and nearby Amundsen Sea Embayment.  observed near/at the grounding zones of this sector as compared to neighboring regions [Bingham et al., 2012, Figure 3], in addition to the lack of an ice shelf to provide a backstress to grounded ice. As such, the general absence of ice shelves throughout this region further signals potentially longstanding ingress of CDW to the bay [Holland et al., 2010]. It is also notable that GL retreat here has occurred at a greater rate over the last decade (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) than previously (Figure 2), a trend which is in general agreement with those reported by Rignot et al. [2008], Pritchard et al. [2009], McMillan et al. [2014, and Wouters et al. [2015] with respect to other indicators of accelerating glacial response over this period. Moreover, following the approach of Park et al. [2013], GL thinning estimates obtained from our 2010-2015 GL retreat rates are consistent with estimates of inland thinning produced through CryoSat-2 swath processing over the same observational period ( Figure S4 and Text S3). Analysis of CryoSat-2-derived rates of surface elevation change reveals extensive thinning inland of the Fox and Ferrigno GLs (maximum À6.40 m yr À1 , σ = ± 0.49 m yr À1 ), which compares well with our Landsat-derived theoretical thinning estimations (mean = À5.96 m yr À1 , σ = ± 3.40 m yr À1 ; Figure S4 and Text S3). There is therefore mounting evidence that the Eltanin Bay component of the BSS is not only prone to marine instability but is already experiencing the early stages of retreat, like nearby Pine Island Glacier [cf. Jenkins et al., 2010].
Ice flowing into Stange Ice Shelf is also remarkable for experiencing extensive GL retreat since 1975 (Figures 2 and S3). The pervasiveness and magnitude of GL retreat there suggest that the ice shelf's capacity to buttress the contributing glaciers' flow has been diminishing as a likely consequence of ice-shelf thinning. Thinning across Stange Ice Shelf has been inferred from radar altimetry between 1992 and 2008 [Holt et al., 2014] and 1994 to 2012 [Paolo et al., 2015], though the rates have varied considerably with location and time [Holt et al., 2014]. The greatest thinning of Stange Ice Shelf has consistently occurred on its westward flanks [Paolo et al., 2015], but where the western Stange Ice Shelf meets the BSS GL (at Berg Ice Stream, Figure 1), GL retreat between 1975 and 2015 has been suppressed relative to farther east (Figures 1 and 2). An explanation for this

Geophysical Research Letters
10.1002/2016GL068972 apparent discrepancy lies in the geometry of the ice bed at the relative locations: a radar profile approximately following the GL along Stange Ice Shelf hints at a shallower bed, or bed protuberance, underlying Berg Ice Stream's GL compared with at the more eastern outlets feeding the ice shelf (notably Lidke Ice Stream; Figure 3). With respect to the temporal changes in GL retreat, the oscillatory nature of the observed retreat rates (i.e., instances of large-to-small-to-large retreat through time) corresponds with the suppression of ice-shelf thinning Paolo et al. [2015] measured for all BSS ice shelves from~2000 to 2008 relative to the periods immediately preceding and succeeding it (i.e., 1994-2000 and 2008-2012). That the GL retreat and ice-shelf thinning trends follow a similar oscillatory pattern lends credence to the hypothesis that regional changes on and around Stange Ice Shelf are, as with Ferrigno Ice Stream, fundamentally driven by persistent CDW access to the ice-shelf cavity [cf. Holt et al., 2014]. Though the sub-shelf and near-ice-front bathymetry of Stange Ice Shelf remain poorly surveyed [Graham et al., 2011], the observed behavior of the GL retreat is strongly suggestive that the ice shelf is underlain by a north-south-aligned deep bathymetric trough that allows CDW to flow from the Ronne Entrance beneath Stange Ice Shelf to the GL. GL retreat observed along the margins of the Abbot and Cosgrove Ice Shelves, as well as on southern Thurston Island, has occurred in locations farther from known regions of CDW ingress (Figure 1). However, most of the larger retreats in the western Bellingshausen Sea (e.g., Rosanova Glacier, Figure 1) took place where Bedmap2 [Fretwell et al., 2013] indicates deeply bedded grounding zones. Similar to Stange Ice Shelf, parts of this coastline exhibited some oscillatory behavior, with the margin between 100°W and 94°W (including Rosanova Glacier) and 93°W and 89°W (eastern Abbot Ice Shelf) experiencing higher retreat in 1985-1990 relative to any subsequent epoch until 2010-2015, when further enhanced GL retreat was observed. Given these observations, we hypothesize that CDW-forced dynamic thinning is also occurring in many of these locations. The dearth of bathymetric data in the western BSS [Graham et al., 2011] leaves open the possibility that throughout the length of the Abbot Ice Shelf, there may be undiscovered access routes for CDW to reach GLs here. Indeed, there is compelling evidence from recently acquired aerogeophysical data that a tectonically rifted basin underlies the majority of the ice shelf [Cochran et al., 2014] and that deep topographic lows may be present near the grounding lines of the eastern Abbot Ice Shelf (Figure 3).
The few cases of GL advance we have observed across the BSS are located where ice at the GL is thin and slow, and the bed steepens/shoals inland (Figure 1). For example, the small advances at King Peninsula ( Figure 1 and segments 59, 61, and 62 in Data Set S1) occurred where ice thickness is~200-400 m and  [Leuschen et al., 2015a[Leuschen et al., , 2015b; gravity data were acquired on 25 October 2012 (flight lines in green) [Cochran and Bell, 2014]. E. Abb. I.S. denotes eastern Abbot Ice Shelf. Inset shows enlarged radar-derived ice bottom depths at Stange Ice Shelf. All other site labels as per Figure 1. Black lines delineate the 2008/2009 MOA GL [Haran et al., 2014].

Geophysical Research Letters
10.1002/2016GL068972 modeled depth-averaged ice velocities are < 100 m yr À1 . In these locations, we hypothesize that the grounding zone geometry is relatively immune to dynamic thinning and that the GL subsequently advanced in response to the high rates of accumulation experienced by the BSS [Lenaerts et al., 2012].
Finally, we highlight the minor GL retreat experienced by the ice streams feeding Venable Ice Shelf, which appears at odds with recent observations that this ice shelf has experienced significant thinning since the 1990s [e.g., Pritchard et al., 2012;Paolo et al., 2015], likely forced by CDW accessing the ice shelf cavity via Belgica Trough (see Figure 1) [Graham et al., 2011]. Indeed, Paolo et al. [2015] report Venable Ice Shelf as having experienced the most dramatic thickness reduction of all Antarctic ice shelves over the period 1994-2012, thinning at a mean rate of 36.1 ± 4.4 m per decade to 82% of its 1994 thickness. Despite this exceptional thinning and known deep bathymetry to the ice shelf's northern margin that potentially connects to the CDW-flooded Belgica Trough, we have observed only limited GL retreat at this location. Here the largest GL change occurred between 1985 and 1990 (0.22 km of retreat at 45 m yr À1 ; segments 30-32 in Data Set S1); and little change in GL position has been observed since. As noted previously, a small retreat is apparent in 2010-2015, but its magnitude (total 0.11 km) is insufficient to support a case that it represents a significant upward or persistent trend in retreat rate [cf. Wouters et al., 2015]. We propose that the discrepancy between oceanic forcing and suppressed GL response observed over the last two decades is a result of the specific bed geometry at/around the grounding zone, presently acting to pin the GL to its current location. Radar data acquired by NASA Operation IceBridge suggest that the bed along the Venable Ice Shelf GL is relatively shallow, including where Wiesnet and Williams Ice Streams feed into the ice shelf ( Figure 3). Significantly, however, where the same surveys profiled ice thickness inland, they partly captured a deep basin underlying Williams Ice Stream from~40 km inland. A subsequent airborne gravity survey of this basin, reaching a farther 50 km inland (Figure 3), showed free-air gravity anomalies 80-100 mGal lower than the surroundings, comparable to those observed over Ferrigno Ice Stream [Bingham et al., 2012], indicating the likely existence of a significant, deeply bedded basin lying beneath Williams Ice Stream. This feature is not identified in Bedmap2 [Fretwell et al., 2013], which did not incorporate these gravity-anomaly data, and renders the setting analogous to both Ferrigno Ice Stream and Pine Island Glacier, which are already experiencing active dynamic thinning. Moreover, in the case of Pine Island Glacier, the current trends are a likely response to the floating section having become ungrounded from a previously longstanding pinning point . We propose Williams Ice Stream, therefore, as an analogous setting with potential for significant future retreat once the coastal pinning point is breached. This hypothesis calls for much improved knowledge of the subglacial geometry of the GL at Williams Ice Stream/Venable Ice Shelf as an important scientific objective to facilitate numerical modeling of the region's future.

Conclusions
Our analysis of Landsat-derived grounding line change reveals that~65% of the grounding line of the Bellingshausen Sea margin retreated between 1990 and 2015. The changes are pronounced, as expected, in locations where known bathymetric lows across the continental shelf facilitate ocean-driven dynamic thinning (e.g., Ferrigno Ice Stream) but have occurred more pervasively along the entire coastline than has previously been reported, implicating the likely ingress of relatively warm circumpolar deep water to the majority of the Bellingshausen Sea margin. Nevertheless, despite extensive thinning over recent decades, we observe only minimal grounding line retreat at Venable Ice Shelf. Hypothesized to be currently pinned to a sub-ice topographic high, future sustained shelf thinning at or exceeding recently reported rates would breach this barrier and facilitate ice retreat into a significant inland basin, analogous to nearby Pine Island Glacier. Together, these findings warrant the requirement for continued observation of this important and dynamically evolving region of Antarctica. acknowledges Geosoft Inc. for their contribution of Oasis Montaj and Geophysics licenses used to process gravity data. The authors also thank the editor and two anonymous reviewers for their insightful comments which improved the manuscript. The authors declare no real or perceived conflicts of interests as a result of this paper. Supplementary data supporting the results of this paper are available in the accompanying supporting information file.