Journal of Geophysical Research: Earth Surface

Widespread acceleration of tidewater glaciers on the Antarctic Peninsula

Authors


Abstract

[1] Over the last half century, the Antarctic Peninsula (AP) has been among the most rapidly warming regions on Earth. This has led to increased summer snowmelt, loss of ice shelves, and retreat of 87% of marine and tidewater glacier fronts. Tidewater-glacier flow is sensitive to changes in basal water supply and to thinning of the terminus, and faster flow leads directly to sea level rise. The flow rates of most AP tidewater glaciers have never been measured, however, and hence their dynamic response to the recent changes is unknown. We present repeated flow rate measurements from over 300 glaciers on the AP west coast through nine summers from 1992 to 2005. We show that the flow rate increased by ∼12% on average and that this trend is greater than the seasonal variability in flow rate. We attribute this widespread acceleration trend not to meltwater-enhanced lubrication or increased snowfall but to a dynamic response to frontal thinning. We estimate that as a result, the annual sea level contribution from this region has increased by 0.047 ± 0.011 mm between 1993 and 2003. This contribution, together with previous studies that assessed increased runoff from the area and acceleration of glaciers resulting from the removal of ice shelves, implies a combined AP contribution of 0.16 ± 0.06 mm yr−1. This is comparable to the contribution from Alaskan glaciers, and combined with estimated mass loss from West Antarctica, is probably large enough to outweigh mass gains in East Antarctica and to make the total Antarctic sea level contribution positive.

1. Introduction

[2] Glacier wastage has contributed approximately 50% of sea level rise over the last century [Church et al., 2001], with about half of this contribution coming from glaciers not part of the polar ice sheets. This loss is primarily as a result of increased summer melting [Dyurgerov, 2002]. Recent results suggest, however, that acceleration of glacier flow, particularly in polar and subpolar regions, is becoming at least as important a contribution to sea level rise as the runoff of surface meltwater. Degree-day models indicate that dynamic effects must account for half of the widespread thinning (up to 10 m yr−1) observed in coastal Greenland over the last decade [Krabill et al., 2004] and this is now supported by new measurements that reveal considerable ongoing retreat and acceleration of numerous Greenland outlet glaciers [Thomas et al., 2003; Howat et al., 2005; Luckman et al., 2006; Rignot and Kanagaratnam, 2006]. This acceleration is most likely due to one or both of: (1) a dynamic response to enhanced meltwater lubrication [Thomas et al., 2000; Zwally et al., 2002; Thomas et al., 2003; Krabill et al., 2004; Rignot and Kanagaratnam, 2006] and (2) a dynamic response to thinning or retreat of tidewater glacier margins [Thomas et al., 2000; Arendt et al., 2002; Rignot et al., 2002; Thomas et al., 2003; Joughin et al., 2004; Krabill et al., 2004; Thomas et al., 2004; Howat et al., 2005; Rignot and Kanagaratnam, 2006]. Whichever process dominates, the rapidity and scale of the observed acceleration raises the prospect that ice sheet and glacier wastage, and hence sea level rise could occur substantially faster and less predictably than was previously envisaged [Intergovernmental Panel on Climate Change, 2001]. Historical records of flow in the high latitudes are, however, particularly scarce and this limits our understanding and capacity to predict the sensitivity of glacier dynamics to change.

[3] This study focuses on the heavily glacierized northern part of the Antarctic Peninsula, extending from 64°S to 70°S (Figure 1). This comprises only 1% (120,000 km2) of the entire Antarctic continent but receives 10% of its snowfall [after van Lipzig et al., 2004]. In contrast to the large continental Antarctic ice sheet, the 1100-km-long northern peninsula is a narrow mountain chain with a maritime climate, drained by over 400 steep, fast-flowing and heavily crevassed glaciers. With a third of its area lying within 200 m of sea level and with summer temperatures frequently above 0°C, this is the only Antarctic region that experiences substantial summer melt (we define summer loosely as the period between first and last melt event), and 80% of its area is classed as a percolation zone [Rau and Braun, 2002]. So little is known about most of the glaciers on the Antarctic Peninsula that they have rarely been included in mountain glacier inventories [Dyurgerov, 2002], but there is clear evidence that glaciers in this area are being affected by regional climatic change.

Figure 1.

Map of the study area in the northern Antarctic Peninsula. Boxes show satellite image frame locations (numbered according to European Space Agency convention). Glacier flow profiles are shown in red. Yellow stars show weather-recording stations used in this study (O'Higgins (OH), Faraday/Vernadsky (F/V), and Rothera (R)). Labels A to H refer to glaciers in Figures 7 and 8.

[4] The high rates of snowfall and widespread melt suggest that Antarctic Peninsula glaciers should be particularly sensitive to climate and there is now ample evidence that since 1950, climate change in this area has been substantial. Strong atmospheric warming of nearly 3°C associated with increased cyclonic circulation [Meredith and King, 2005] has increased the duration and intensity of summer melting, measured as positive-degree-days (PDDs), by up to 74% [Vaughan, 2006]. It is generally accepted that ice shelves on both the east and west coasts that have experienced progressive retreat and then abrupt collapse have done so as summer meltwater production rose to cross a threshold of ice shelf stability [Morris and Vaughan, 2003]. As a consequence, tributary glaciers accelerated and now contribute an extra ∼0.07 mm yr−1 to global sea level rise [Rott et al., 1996; Rignot et al., 2004; Scambos et al., 2004; Rignot et al., 2005]. The near-surface ocean has warmed by more than 1°C and wind fields have changed, reducing sea-ice production and persistence [Smith and Stammerjohn, 2001; Meredith and King, 2005; Harangozo, 2006]. In addition, the number of recorded precipitation events has increased by 50% [Turner et al., 1997] and snowfall by 20% in at least one location [Schneider, 1999]. Finally, since the 1940s, 87% of non-ice-shelf glacier fronts have retreated [Cook et al., 2005].

[5] The dynamic response of the Antarctic Peninsula's many tidewater glaciers to this near-ubiquitous retreat of their fronts, increased snowfall and increased summer melting remains unknown, however, and its impact has not been accounted for in sea level predictions. In this study, we calculate the change in flow rate of these glaciers, discuss which environmental changes are driving those changes and reassess the current sea level contribution of the region.

2. Method

[6] Several aspects of the study area dictate a remote-sensing approach to monitoring the flow of the region's glaciers. The scale, steep topography and difficult surface conditions mean that most are inaccessible to field workers. Secondly, there is strong evidence that the behavior of individual glaciers may be quite different from the mean [see Cook et al., 2005] and so the use of a few benchmark glaciers could be misleading. Here we have used Synthetic Aperture Radar (SAR) feature tracking on 75 images to measure change in glacier flow rate on a regional scale.

[7] Directly comparable SAR data exist during the European Space Agency ERS satellite missions from 1992 onward, though these were typically limited to pairs of images spanning a single 5-week interval in the period from late December to early February (southern hemisphere summer) because a summer-only Antarctic station, O'Higgins (Chile), was used as the regional satellite download station. For three areas, we were, however, able to make repeated measurements within four summer seasons (Table 1). These data were crucial since they allowed us to investigate the variation in flow rate associated with the arrival of the melt season, through midsummer and from mid to late summer. In total, velocity profiles were measured along the approximate centerlines of 337 glaciers in at least 5 summers, and a subset of these was measured in as many as nine summers, between 1992 and 2005 (Figure 1; listed in auxiliary material).

Table 1. Direct Comparison of Mean Flow Rates Within Melt Seasonsa
FrameSeasonFlow Measurement PeriodMean, m d−1
  • a

    None of the seasonal changes are significant (outside the standard error).

4923early summer midsummer8 Dec to 12 Jan 1992/1993 12 Jan to 16 Feb 19930.77 ± 0.02 0.78 ± 0.02
winter early summer midsummer15 Oct to 19 Nov 2004 19 Nov to 24 Dec 2004 24 Dec to 28 Jan 2004/20050.79 ± 0.02 0.77 ± 0.02 0.78 ± 0.02
 
4941 and 4959early summer midsummer21 Nov to 26 Dec 2002 26 Dec to 30 Jan 2002/20030.77 ± 0.01 0.76 ± 0.01
 
5841midsummer late summer21 Jan to 16 Feb 1994 16 Feb to 19 Mar 19940.87 ± 0.02 0.86 ± 0.02

[8] SAR feature tracking is well suited to highly crevassed, fast-flowing mountain glaciers typical of the majority of those in the study area [e.g., Pritchard et al., 2005]. It differs slightly from SAR speckle tracking in that it is used to track robust physical features rather than speckle, the ephemeral product of backscatter interference from an ensemble of subpixel-scale scatterers. It is similar to optical-image feature tracking except that SAR is more sensitive to snow and ice surface characteristics and is not affected by cloud cover, darkness or sensor saturation in the way that optical images are, particularly over the Antarctic Peninsula. The SAR feature-tracking technique has been used successfully to measure glacier velocity in a number of alpine environments but here we have introduced modifications to suit the specific conditions found on the Antarctic Peninsula.

[9] The offset of features such as crevasses tracked between pairs of satellite images is made up of contributions from: (1) slight differences in satellite position during image acquisition and (2) ice movement between the acquisition times. The satellite position can vary both along the track of the orbits (azimuth direction) and across track (range direction) by typically tens to hundreds of meters. To remove this orbital contribution, automatically retrieved offsets of static rock areas were used to define orbit polynomials [after Pritchard et al., 2005]. This is a relatively simple process over flat terrain, but on the Antarctic Peninsula, the characteristic SAR image distortion of the high-relief mountain peaks and ridges introduced systematic error into the range-direction orbit polynomial. This is because the extent of relief distortion varies when viewed from differing points in space. Correction of this distortion was not possible owing to the low quality of digital elevation models available for this area. Such systematic error in the orbits transfers directly into systematic error in glacier flow rates in the range direction that we estimate could reach 0.2 m d−1. Furthermore, SAR pixels have a five-to-one range-to-azimuth aspect ratio on the ground, hence the precision of feature tracking in the range direction is relatively crude compared to that in the azimuth direction.

[10] To study potentially small, systematic changes in glacier flow across large areas, it was important that we minimized systematic error (inaccuracy) and random error (imprecision) in the tracking measurements. To improve accuracy, we first refined the orbit polynomials for each image pair using samples of the image offsets that we selected manually from low-relief rock features all of similar altitude (i.e., with similar relief distortion). We assessed improvement in the orbit polynomial by looking at the residual apparent motion of other static areas after orbit removal. We repeated this process until no significant improvement was made. This reduced the systematic error in both range and azimuth, but it remained relatively large in the range direction (> ±0.1 m d−1). Consequently, we chose to reject the range component of all glacier flow measurements and made all comparisons of flow using only the azimuth component. The azimuth component has lower systematic and random error due to the absence of topographic distortion and finer pixel spacing, and hence is better suited to detecting small and widespread changes in glacier flow rate. The absolute systematic error in each azimuth data set was reduced to typically <±0.02 m d−1, and the random error to approximately ±0.06 m d−1 (sample sizes 13,000 to 60,000).

[11] Our use of the single, azimuth-only component of the flow meant that we could only make comparisons between image frames collected along the same direction of travel (ascending or descending orbits). In most cases, images were acquired from the same orbital tracks, although in 1997 and 1998, they were acquired from the adjacent track with a two-thirds image overlap. This introduces a rotation in the azimuth direction of the images of ∼0.6° relative to the other years, and hence rotation of the measured glacier flow component. The systematic error introduced into the flow measurement is, however, negligible.

[12] After refinement, the azimuth-only flow measurements were filtered for outliers to remove false-match tracking offsets and lightly smoothed with a linear distance-weighted 3 × 3 averaging kernel before being coregistered. The low quality and resolution of existing digital elevation models of the Antarctic Peninsula precluded reprojection of flow downslope and so this lack of topographic correction introduces errors when comparisons are made between data sets acquired from adjacent tracks. In this case though, the low relief of the glacier surfaces and the large track overlap produce flow errors of <±0.5% (i.e., only millimeters or less per day).

[13] Once coregistered, we extracted measurements of flow along defined profiles on each of the glaciers for comparison within seasons and between years (Figure 1). Because of the technique used and the requirement for trackable surface features, the flow measurements came primarily from the lower, more crevassed reaches of glaciers. In all cases, the profiles are discontinuous: gaps exist where no features could be tracked and these gaps vary between data sets. For analysis, we took a master set of profiles along which we extract the flow values, but the locations of these profiles are necessarily a compromise aimed at maximizing coverage in numerous data sets. In some cases, for example when comparing specific pairs of flow data sets within the same season, customized profiles were defined that maximized the number of measurements available for comparison.

3. Analysis

[14] In this section, we present results that allow us to compare flow measurements both within and between summers. We look for changes in flow rate that could be driven by the seasonal progression of summer melt, and also changes that persist between years and could be indicative of sustained trends in glacier dynamics in the study area.

3.1. Flow Rate Changes and Summer Melt

3.1.1. Seasonal Changes in Flow Rate

[15] The flow rate of alpine and subpolar glaciers has frequently been observed to vary with the rate and quantity of meltwater production and with the stage of the melt season [Bindschadler, 1983; Meier and Post, 1987; Willis, 1995], but the correlation is usually not a simple one. Early in the melt season, water is often impeded by an inefficient basal drainage system that cannot drain rapid increases in water supply. This leads to high basal water pressure, reduced basal friction and faster glacier flow. As the melt season progresses the drainage system evolves to become more efficient, reducing the sensitivity of glacier flow to water input [Willis, 1995]. When comparing glacier flow rates between summers, it is therefore necessary to establish the sensitivity of the glacier flow to the intensity of meltwater supply and to timing within the melt season.

3.1.1.1. Timing Within the Melt Season

[16] In four of the summers in this study (1993, 1994, 2003 and 2005), more than one pair of images were available for velocity measurements (listed in auxiliary material). This allowed us to look at the magnitude of changes in flow rate as the melt season began and progressed. In the absence of direct measurements of surface melt, we counted positive-degree-days (PDDs) using temperatures measured at local meteorological stations to estimate the melt intensity in our study area [cf. Vaughan, 2006]. For each 35-day flow measurement period, we added the PDDs for the 35 days spanning the measurement plus the preceding 14 days in an attempt to account for all relevant melt events. We then related the flow variations in each frame to the variations in PDDs at the closest research station, Rothera, Faraday/Vernadsky or O'Higgins (Figure 1). We did not interpolate or attempt to model variation in PDDs between stations because they are known to be highly correlated [Vaughan, 2006].

[17] In frame 4923 (Figure 1), two measurements in early to mid summer 1992−1993 and three spanning the winter to midsummer transition in 2004−2005 were possible. We found that the change in flow rate within each season was smaller than the interannual change and smaller than the overall change over the 12-year period (Figure 2a). Similar results emerged for frames 4941, 4959 (Figure 2b) and 5841 (Figure 2c), in which changes in flow rate from mid to late summer 1994 and early to mid summer 2002−2003 were substantially smaller than the interannual changes.

Figure 2.

Change in mean flow rate (black diamonds) and PDDs (gray circles) for (a) frame 4923 (relative to 1993, PDD total = 48), (b) frame 4941 and 4959 (relative to 1993, PDD total = 70), and (c) frame 5841 (relative to 1992, PDD total = 41) (Figure 1). Arrows indicate summers with multiple measurements. Error bars are standard errors. Frame locations are shown in Figure 1.

[18] Figure 2 shows a comparison of the mean flow rates of several years relative to a base year (1992 or 1993). The most direct test for changes in flow within any one season is to compare these intraseasonal data sets directly to each other, increasing the number of sample points available for comparison. When we did this, we noted that in fact, no significant change in mean flow rate occurred within any of the summers studied as the melt season progressed (Table 1). These results establish that between four different melt seasons in four frames, flow varied significantly but that this variation was not dominated, or even substantially influenced, by the timing within a given melt season.

3.1.1.2. Summer Melt Intensity

[19] In the last section we established that flow rate did not vary according to timing within a melt season, but we can also address the question of how sensitive flow rate is to melt intensity by comparing the flow rates to the number of PDDs that occurred during and immediately prior to the flow rate measurement period.

[20] In the northernmost peninsula (frame 4923), we showed that flow rate in the 1992/3 and 2004/5 seasons varied little (Figure 2a and Table 1). In 1992/1993, the number of PDDs also changed little between measurement periods. In 2004/2005, however, the measurements span the transition from late winter into midsummer and the number of PDDs increased dramatically, but this had no significant effect on flow rate. Farther south (frames 4941 and 4959), the same is true of the 2002/2003 season, with little intraseasonal flow rate variation despite large increases in PDDs through the summer (Figure 2b and Table 1).

[21] Overall, we see no correspondence between high or rapidly increasing PDDs and fast flow rates, and no clear relationship emerges when we increase or decrease the period over which the PDDs are counted. In the study area, flow rate appears to be largely independent of summer melt intensity.

3.1.2. Interannual Change in Flow Rate

[22] For the years 1993, 1996, 1997, 1998 and 2003 we have near complete data coverage over all five frames that comprise the study area, and we extracted flow measurements along the profiles in Figure 1 for each of these years. Significant interannual changes in mean flow rate occurred with an overall acceleration of 10% over the 11-year period (Figure 3). When the flow changes in each frame are plotted separately, we see that similar changes occurred between years through the entire study area, with a trend punctuated by a peak in 1997 (Figure 4).

Figure 3.

Mean flow rate each summer relative to 1993 for the full study area. Error bars show the standard error of the percent change (approximately ±0.2%). The number of samples compared between a given year and 1993 ranges from 13,000 to 16,000.

Figure 4.

Interannual changes in mean flow rate by frame, relative to 1993 for individual frames. Error bars are standard errors. Frame locations are shown in Figure 1.

[23] Comparing the measurements by frame between the years 1993 and 2003 only (with profiles tailored to optimize coverage in just these years), it is clear that the acceleration was widespread across the majority of glaciers and was not dominated by the behavior of a small number of glaciers. This prevalence of an acceleration trend is demonstrated by the proportion of profiles in each frame that showed acceleration (45% to 73%) relative to deceleration (9% to 24%) (Table 2). The results from the optimized 1993–2003 profiles reveal a slightly greater overall acceleration of 12.4 ± 0.9% over the period than was apparent in Figure 3. The mean flow rate for those glaciers in the five frames measured in both 1993 and 2003 rose from 0.86 ± 0.01 m d−1 to 0.97 ± 0.03 m d−1 (error estimates are standard errors). Owing to a lack of data prior to 1993, we do not know when the acceleration trend began, only that it was underway throughout our study period.

Table 2. Percentage Change in Mean Flow Rate for All Points Common to the 1993 and 2003 Profiles, Split Into Framesa
FramePercent Change in Mean Flow Rate, 1993 to 2003Number of ProfilesPercent Profiles Accelerated (by >5%)Percent Profiles Slowed (by >5%)
  • a

    See Figure 1 for frame details. For each frame, the number of tracking measurement points compared ranges from 1800 to 2800.

4923+ 7.8 ± 0.5556020
4941+ 15.2 ± 0.3714517
4959+ 12.6 ± 0.3755924
4977+ 12.7 ± 0.5706713
4995+ 13.4 ± 0.570739

[24] The PDD count over the 49-day period in each of the five summers with full coverage is shown in Figure 5. The count for each station varied between these summer periods in a similar way. In each case the number of PDDs increased with latitude between the stations (i.e., more melt farther south, a counterintuitive result but not unreasonable given the findings of another study [Vaughan, 2006]). There is again a lack of correspondence between fast flow and high melt, most notably in 1997 when flow rates peaked while PDDs were low.

Figure 5.

Positive degree days (PDDs) over the 35-day flow measurement periods plus the preceding 2 weeks at each of the three stations, Rothera in the south (gray circles) and Faraday (black squares) and O'Higgins (triangles) in the north (locations in Figure 1).

[25] As described above, we used PDDs at meteorological stations as a proxy for melt intensity over our study area. We did not interpolate our PDD counts between stations, though this is justified as the station counts are highly correlated [Vaughan, 2006]. The use of PDDs is a simplification because melt sensitivity to temperature is also dependent on many other climate variables (cloudiness, relative humidity, rainfall), and many local conditions (slope, aspect, proximity to rock outcrop) and in particular, it can be three times higher at the northern limit of our study area than the southern limit [Braun and Schneider, 2000]. Furthermore, we did not attempt to calculate runoff, the free water that leaves the snowpack after saturation. However, the lack of a convincing correspondence between flow rate and PDDs on either a seasonal (section 1.1) or interannual timescale (section 1.2) in any of the frames indicates that the flow of the study glaciers is largely insensitive to summer melt intensity and timing within the melt season. This may be because the volume of summer melt remains relatively low (e.g., 100 to 200 PDD per year at Rothera compared, for example, to approximately 330 to 640 PDD per year at similar altitude in western Greenland (Jakobshavn Isbræ) [Thomas et al., 2003]), and is perhaps insufficient to affect basal water pressure significantly. Hence we assert that the observed changes in glacier flow rate are not directly driven by summer melting, nor are they a response to evolving drainage systems through the melt season. They are instead the result of some other, regionally operating, sustained driver of change. The candidates for such a driver are discussed in the following sections.

3.2. Increased Snowfall

[26] Snowfall on the Antarctic Peninsula is estimated to have increased by 20% since the 1950s [Schneider, 1999] and a thickening glacier generates greater driving stress and hence a faster flow rate as it adjusts toward a new mass balance. However, the response to increasing accumulation rates is expected to be slow to operate and to produce a relatively small acceleration. We estimate that for glacier dimensions typical of the western Antarctic Peninsula, even a 10-m thickening over the study period (much larger than the changes actually observed [Morris and Mulvaney, 2004]) would cause an acceleration of less than 2%. We therefore cannot explain the acceleration as a response to increasing accumulation on the glaciers.

3.3. Frontal Thinning and Retreat

[27] On the Antarctic Peninsula, there has been a well-documented, widespread retreat of tidewater glacier fronts that was particularly rapid over the last 15 years [Cook et al., 2005], (Figure 6). Elsewhere, numerous observations have been made of positive relationships between calving rate and near-frontal flow rate in tidewater and marine glaciers [e.g., Meier and Post, 1987]. Furthermore, theoretical models have been constructed to show that changes in near-grounding-line conditions (position and ice thickness) can have impacts on the flow rate of glaciers some considerable distance inland [Payne et al., 2004; Dupont and Alley, 2005]. Together, these observations prompt us to look for a relationship between glacier flow rate and frontal retreat in our sample.

Figure 6.

Change in glacier frontal position for each frame through time, with 95% confidence-interval error bars (this analysis used data presented by Cook et al. [2005]). A total of 282 fronts were used. This shows widespread frontal retreat over the period of this study. The asterisk denotes exclusion of a 4.5-km retreat of Widdowson Glacier from 1997 to 2001 that dominates the statistics for frame 4977.

[28] From the SAR images used here, it is apparent that some glacier fronts in the study area retreated substantially between 1993 and 2003 (eight retreated by several hundred meters). Glaciers with the most marked retreats also accelerated considerably and some examples are given in Figures 7 and 8. The profiles in Figure 8 show a particularly close correspondence between sudden, large-scale retreat and sudden acceleration. In this case, a retreat of 1400 to 1600 m between 1998 and 2003 was associated with an acceleration of 40% for both glaciers. While the majority of the study glaciers experienced much smaller change in front position and more modest acceleration, these specific, extreme examples suggest a link between frontal retreat and acceleration of the Antarctic Peninsula glaciers.

Figure 7.

Long profiles of flow rate for six of the glaciers that retreated the most between 1993 and 2003 (glacier fronts are at left, locations A to F in Figure 1). The bar charts show the retreat history for these glaciers. These graphs demonstrate the quality of the tracking data used and show that glaciers with large retreats also experienced substantial acceleration in the study period, though it is not possible to discern a causal link from these data.

Figure 8.

Long profiles of flow rate and bar charts of frontal retreat for glaciers G and H (location in Figure 1). Inset image shows the tracking flow results from 1993 (sea to south of image) overlaid with the profile lines (white) and the 1993 and 2003 frontal positions (black). Arrow shows principal direction of flow. These glaciers retreated by 1400 to 1600 m over this period, associated with acceleration of 40%.

[29] Recent studies indicate, however, that a high calving rate and retreat is an effect and not the cause of fast flow [van der Veen, 1996; Venteris, 1999; Vieli et al., 2001; van der Veen, 2002]. In those studies, the trigger for glacier retreat is considered to be thinning of the terminus due to an increasingly negative mass balance. Such thinning will inevitably reduce the effective stress at the bed of a tidewater margin, reducing basal resistance and permitting faster sliding. This leads to further thinning until eventually, the terminus floats and rapid calving and retreat begin. The speed of the ensuing retreat is then controlled to a large extent by the slope and depth of the fjord bed: retreat through an over-deepening immediately inland of an existing grounding line can be particularly rapid as the greater water depth promotes fast sliding and hence rapid thinning to flotation [Vieli et al., 2001]. Acceleration can subsequently propagate more slowly up-glacier driven by a steepening of the surface slope as the terminus thins [Payne et al., 2004; Howat et al., 2005].

[30] A plausible explanation for the acceleration of flow seen in this study is that the tidewater glaciers of the Antarctic Peninsula are thinning in their terminal reaches owing to a widespread reduction in mass balance at low altitudes, so that many of these glacier fronts are approaching flotation, experiencing reduced effective pressure and sliding faster. We suggest that in many cases, the fronts progressively reach flotation and then steadily retreat, while in some (the examples of rapid retreat given in Figures 7 and 8), it may be that a threshold was reached during the observation period whereby the fronts retreated into an over-deepened basin in the fjord bed (a common fjord configuration), initiating the observed rapid speed-up and consequent, simultaneous rapid retreat. This mechanism is consistent with the lack of strong retreat in particular in frames 4941 and 4959 (Figure 6) at a time of acceleration (Figure 4 and Table 2).

[31] A similar thinning-driven process appears to be under way in coastal Greenland, where sustained melt-driven and dynamic thinning occurred through the 1990s [Krabill et al., 2000; Thomas et al., 2003; Krabill et al., 2004]. This appears to have led to the near-synchronous, strong acceleration and retreat of numerous outlet glaciers, most notably Helheim, Kangerdlugssuaq and Jakobshavn [Luckman et al., 2006; Rignot and Kanagaratnam, 2006].

[32] A reduction in mass balance at low altitude is supported by measurements showing the down-wasting of glacier margins both inland and coastal [Jamieson and Wager, 1983; Splettoesser, 1992; Morris and Mulvaney, 1995; Smith et al., 1998] (and similar subsequent unpublished data, A. M. Smith, personal communication, 2006), a reduction in seasonal snow cover [Fox and Cooper, 1998] and the drastic, melt-driven loss of ice shelves on both the east and west coasts of the peninsula [e.g., Morris and Vaughan, 2003]. The cause of the mass balance deficit can reasonably be ascribed to the well-documented climatic change experienced by the region over the last half century. This has involved a change to increased cyclonic atmospheric circulation and strong atmospheric and near surface warming of the sea [e.g., Meredith and King, 2005; Turner et al., 2005], and has led to much increased duration and intensity of summer melting [Vaughan, 2006]. The rate of down-wasting measured at one site was found to be strongly correlated to PDDs over a period of several years [Smith et al., 1998].

[33] Assuming that thinning does drive the retreat and acceleration, an order-of-magnitude calculation of thinning in the study area can be made using one suggested thinning-retreat relation (equations (13) and (14) of van der Veen [1996]). For the measured Antarctic Peninsula mean glacier frontal retreat rate of 33 m yr−1 from 1990 to 2004 [after Cook et al., 2005], an assumed mean near-terminus basal slope of 0° (no data available) and a calculated mean terminus surface slope of 9°, the implied mean terminus ice thinning rate is of order 5 m yr−1. This can be compared to tidewater glacier thinning rates in Alaska of up to 30 m yr−1 on Columbia Glacier [Arendt et al., 2002]; in Greenland, 20 m yr−1 on Helheim glacier [Howat et al., 2005], 8−21 m yr−1 on Jakobshavn Isbrae [Thomas et al., 2003] and up to 10 m yr−1 on Kangerdlugssuaq glacier; in Patagonia, 2−10 m yr−1 [Venteris, 1999]; and up to 2 m yr−1 on Fleming Glacier, to the south of the study area on the Antarctic Peninsula west coast [Rignot et al., 2005]. It is, therefore, not unreasonable that the glaciers in the study area could have thinned at a sufficient rate to cause the observed retreat, supporting the proposed model for thinning-driven retreat and acceleration.

3.4. Implications for Sea Level

[34] A recent assessment of data collected between the mid-1950s and mid-1990s, showed that Alaskan glaciers covering an area of 90 000 km2 were thinning and retreating sufficiently rapidly to be adding to global sea level rise at a rate of 0.14 ± 0.04 mm yr−1 (probably increasing to 0.27 ± 0.10 mm yr−1 in the late 1990s), the largest estimated glaciological contribution to rising sea level to date [Arendt et al., 2002]. Since the regional climate warming on the Antarctic Peninsula is likely to have exceeded that of Alaska over the past few decades [Folland et al., 2001] and the climatological and glaciological setting of these subpolar alpine glacier systems is quite similar, we are prompted to draw a comparison. With similar rates of climate change, area, and snowfall, we might expect the Antarctic Peninsula to be making a contribution to sea level rise that is comparable to that from Alaska.

[35] The tidewater/marine glacier systems on the grounded Antarctic Peninsula (north of 70°S and excluding ice shelves and the former tributary glaciers of Larsen A, B and Wordie shelves) have an area of 95 000 km2 and a mean net annual accumulation of 143 ± 29 Gt yr−1 [after van Lipzig et al., 2004] (we assign a 20% uncertainty based on the mean absolute bias between modeled and observed mass balance [after van Lipzig et al., 2004]). Assuming that this area was in balance in 1993 and that the measured 12% acceleration is representative of an increase in annual flux, we estimate that this acceleration in flow has caused an increase in annual sea level contribution from the study area of 0.047 ± 0.011 mm from 1993 to 2003. Many of these glaciers are known to have been in retreat for some time before 1993, hence the true imbalance resulting from this proposed dynamic response to climate change is likely to be larger. This 0.047 mm yr−1 contribution can be added to the 0.032 ± 0.024 mm yr−1 increased-runoff contribution calculated for 2000 [Vaughan, 2006]; the 0.07 ± 0.02 mm yr−1 contribution due to the speed up of former Larsen B Ice Shelf tributaries and Larsen A's Drygalski Glacier after the loss of those ice shelves [Rignot et al., 2004] (we assign a 30% uncertainty based on ∼10% uncertainty in balance flux and ∼20% in subsequent flux measurements [after Rignot et al., 2004]); and 0.007 ± 0.003 mm yr−1 contribution due to speed up of the Wordie Ice Shelf tributary, Fleming Glacier [Rignot et al., 2005]. These figures suggest that circa 2005, the Antarctic Peninsula was contributing to global sea level rise through enhanced melt and glacier acceleration at a rate of 0.16 ± 0.06 mm yr−1 (this can be compared to an estimated total Antarctic Peninsula ice volume (95,200 km3) equivalent to 242 mm of sea level).

[36] At present, this northerly section of the Antarctic Peninsula is not resolved in satellite radar altimeter assessments of ice sheet thickness change [e.g., Wingham et al., 2006], and so we cannot estimate how much of this increased outflow is balanced by increased accumulation. However, one estimate of mass change due primarily to temperature-dependent increases in snowfall on the peninsula produces a contribution to sea level of approximately −0.003 mm yr−1 [Morris and Mulvaney, 2004].

4. Conclusions

[37] The northwestern section of the Antarctic Peninsula on which we have focused in this study contains more than 300 glaciers whose lower reaches are subject to considerable summer melting, rising summer temperatures and a general retreat of ice fronts. These glaciers cover an area of around 95,000 km2 and this area receives 8% of all Antarctic snowfall, so any imbalance between snowfall and glacier flux could imply a significant contribution to sea level rise.

[38] We have presented evidence that the average flow rate of glaciers on the west coast of the peninsula increased by 12% from 1993 to 2003. This increase was part of an acceleration trend that was similar in each of 5 adjacent frames spanning 63°30′S to 67°30′S, with 60% of glaciers accelerating by more than 5%. The interannual changes in flow rate were larger than the changes measured through the melt seasons. This indicates that the trend is not an artifact of measuring the flow in different stages of the melt season. Furthermore, the interannual and seasonal changes in flow do not correspond to the intensity of summer melting (the number of PDDs counted at local meteorological stations). This indicates that the acceleration signal is not the result of variations in summer weather. Furthermore, we do not believe that increased accumulation can account for the observed acceleration.

[39] The acceleration appears to be linked to frontal retreat and thinning. The pattern of most widespread acceleration in the south corresponds to an area of increasing frontal retreat, and the most prominent cases of strong acceleration were on glaciers that showed strong retreat observed over the study period. We suggest that thinning of the buoyant glacier termini under a prolonged negative mass balance regime can explain both the acceleration and retreat [after van der Veen, 1996; Vieli et al., 2001]. In this model, faster sliding results as the terminus thins and effective pressure falls, and this feeds back into further thinning. The calving rate increases as the margin goes afloat, and where the increased calving exceeds the increased flux, the glacier retreats. In broad terms, this is in agreement with models and observations of tidewater systems elsewhere [Luckman et al., 2006; Rignot and Kanagaratnam, 2006]. Given the regional scale of the response, it seems likely that the thinning was triggered by the observed warming and consequent increase in summer melt observed on the Antarctic Peninsula since the 1950s [Vaughan, 2006].

[40] We estimate that the net sea level contribution of the Antarctic Peninsula circa 2005 was 0.16 ± 0.06 mm yr−1. This is of similar magnitude to the contribution from Alaskan glacier retreat [Arendt et al., 2002]. Furthermore, this contribution is larger than the 0.07 mm yr−1 [Rignot and Thomas, 2002] or 0.08 ± 0.03 mm yr−1 [Zwally et al., 2005] or −0.08 ± 0.08 mm yr−1 [Wingham et al., 2006] calculated for the West and East Antarctic Ice Sheets combined, which excluded the northern Antarctic Peninsula. It is probable, therefore, that the sea level contribution identified in this study is large enough to make the total Antarctic sea level contribution positive, regardless of which of the above values is chosen.

[41] On this basis, the glaciers of the Antarctic Peninsula undoubtedly warrant further monitoring, particularly given that the changes are apparently a rather complex response to relatively modest rates of summer warming, rates that could be sustained for decades to come. The pressing requirement to build a predictive capacity for this area, explicit from the rest of the Antarctic ice sheet, is clear.

Acknowledgments

[42] We thank Alison Cook and Jane Ferrigno for access to their database of glacier fluctuations and Peter Fretwell for AP ice volume calculations. Satellite imagery was made available from European Space Agency through the Vectra initiative. We thank Hilmar Gudmundsson, Robert Arthern, and Eric Rignot for helpful discussions and Ted Scambos, Gordon Hamilton, and Tavi Murray for suggesting valuable improvements to the manuscript.

Ancillary