Volume loss from Antarctic ice shelves is accelerating

See allHide authors and affiliations

Science  17 Apr 2015:
Vol. 348, Issue 6232, pp. 327-331
DOI: 10.1126/science.aaa0940

Disappearing faster around the edges

The floating ice shelves around Antarctica, which buttress ice streams from the continent and slow their discharge into the sea, are thinning at faster rates. Paolo et al. present satellite data showing that ice shelves in many regions around the edge of the continent are losing mass. This result increases concern about how fast sea level might rise as climate continues to warm. If warming continues to cause ice shelves to thin, as they have for the past couple of decades, their disappearance may allow land-based ice to collapse and melt.

Science, this issue p. 327


The floating ice shelves surrounding the Antarctic Ice Sheet restrain the grounded ice-sheet flow. Thinning of an ice shelf reduces this effect, leading to an increase in ice discharge to the ocean. Using 18 years of continuous satellite radar altimeter observations, we have computed decadal-scale changes in ice-shelf thickness around the Antarctic continent. Overall, average ice-shelf volume change accelerated from negligible loss at 25 ± 64 cubic kilometers per year for 1994–2003 to rapid loss of 310 ± 74 cubic kilometers per year for 2003–2012. West Antarctic losses increased by ~70% in the past decade, and earlier volume gain by East Antarctic ice shelves ceased. In the Amundsen and Bellingshausen regions, some ice shelves have lost up to 18% of their thickness in less than two decades.

The Antarctic Ice Sheet gains mass through snowfall and loses mass at its margin through submarine melting and iceberg calving. These losses occur primarily from ice shelves, the floating extensions of the ice sheet. Antarctica’s grounded-ice loss has increased over the past two decades (1, 2), with the most rapid losses being along the Amundsen Sea coast (3) concurrent with substantial thinning of adjoining ice shelves (4, 5) and along the Antarctic Peninsula after ice-shelf disintegration events (6). Ice shelves restrain (“buttress”) the flow of the grounded ice through drag forces at the ice-rock boundary, including lateral stresses at sidewalls and basal stresses where the ice shelf rests on topographic highs (7, 8). Reductions in ice-shelf thickness reduce these stresses, leading to a speed-up of ice discharge. If the boundary between the floating ice shelf and the grounded ice (the grounding line) is situated on a retrograde bed (sloping downwards inland), this process leads to faster rates of ice flow, with potential for a self-sustaining retreat (7, 9, 10).

Changes in ice-shelf thickness and extent have primarily been attributed to varying atmospheric and oceanic conditions (11, 12). Observing ice-shelf thickness variability can help identify the principal processes influencing how changing large-scale climate affects global sea level through the effects of buttressing on the Antarctic Ice Sheet. The only practical way to map and monitor ice-shelf thickness for this vast and remote ice sheet at the known space and time scales of ice-shelf variability is with satellite altimetry. Previous studies have reported trends based on simple line fits to time series of ice-shelf thickness (or height) averaged over entire ice shelves or broad regions (4, 13) or for short (~5-year) time intervals (5, 14, 15). Here, we present a record of ice-shelf thickness that is highly resolved in time (~3 months) and space (~30 km), using the longest available record from three consecutive overlapping satellite radar altimeter missions (ERS-1, 1992–1996; ERS-2, 1995–2003; and Envisat, 2002–2012) spanning 18 years from 1994 to 2012.

Our technique for ice-shelf thickness change detection is based on crossover analysis of satellite radar altimeter data, in which time-separated height estimates are differenced at orbit intersections (13, 16, 17). To cross-calibrate measurements from the different satellite altimeters, we used the roughly 1-year overlap between consecutive missions. The signal-to-noise ratio of altimeter-derived height differences for floating ice in hydrostatic equilibrium is roughly an order of magnitude smaller than over grounded ice, requiring additional data averaging to obtain comparable statistical significance. We aggregated observations in time (3-month bins) and space (~30-km cells). Because the spatial distribution of crossovers changes with time (due, for example, to nonexact repeat tracks and nadir mispointing), we constructed several records at each cell location and stacked them in order to produce a mean time series with reduced statistical error (18). We converted our height-change time series and rates to thickness changes by assuming that observed losses occurred predominantly at the density of solid ice (basal melting) (4, 5, 17). This is further justified by the relative insensitivity of radar measurements to fluctuations in surface mass balance (18). For volume changes, we tracked the minimum (fixed) area of each ice shelf (18). We assessed uncertainties for all estimates using the bootstrap approach (resampling with replacement of the residuals of the fit) (19), which allows estimation of formal confidence intervals. All our uncertainties are stated at the 95% confidence level [discussion of uncertainties are provided in (18) and the several corrections applied are stated in (20)].

We estimated 18-year trends in ice-shelf thickness by fitting low-order polynomials (degree n ≤ 3) to the data using a combination of lasso regularized-regression (21) and cross-validation for model-parameter selection (the shape of the fit is determined by the data). This combined approach allowed us to minimize the effect of short-term variability on the 18-year trends. Relative to previous studies (4, 5, 13, 22), we have improved estimations by (i) using 18-year continuous records, (ii) implementing a time series averaging scheme so as to enhance the signal-to-noise ratio, and (iii) using a robust approach to trend extraction.

The 18-year average rate of thickness change varies spatially (Fig. 1). On shorter time scales, trends are highly variable but spatially coherent (Fig. 2 and movie S1). We divided our data set into eight regions on the basis of spatial coherence of long-term ice-shelf behavior and calculated time series of ice-shelf thickness change (relative to series mean) for each region (Fig. 3). The largest regional thickness losses were in the Amundsen and Bellingshausen seas, with average (and maximum) thinning rates of 19.4 ± 1.9 (66.5 ± 9.0) m/decade and 7.4 ± 0.9 (64.4 ± 4.9) m/decade, respectively. These values correspond to ~8 and 5% of thickness loss over the 18 years for the two regions, respectively. These two regions account for less than 20% of the total West Antarctic ice-shelf area but, combined, contribute more than 85% of the total ice-shelf volume loss from West Antarctica. The area-averaged time records of ice-shelf thickness and volume for the West and East Antarctic sectors (Fig. 1, bottom left), broad regions (Fig. 3), and single ice shelves (fig. S1) at 3-month time intervals show a wide range of temporal responses with large interannual-to-decadal fluctuations, stressing the importance of long records for determining the long-term state of the ice shelves. Comparing our long records with simple linear trends obtained for the periods of single satellite missions [such as the 5-year ICESat time span used in (5)] shows that it is often not possible to capture the persistent signals in the shorter records (Fig. 3 and fig. S1).

Fig. 1 Eighteen years of change in thickness and volume of Antarctic ice shelves.

Rates of thickness change (meters per decade) are color-coded from –25 (thinning) to +10 (thickening). Circles represent percentage of thickness lost (red) or gained (blue) in 18 years. Only significant values at the 95% confidence level are plotted (table S1). (Bottom left) Time series and polynomial fit of average volume change (cubic kilometers) from 1994 to 2012 for the West (in red) and East (in blue) Antarctic ice shelves. The black curve is the polynomial fit for All Antarctic ice shelves. We divided Antarctica into eight regions (Fig. 3), which are labeled and delimited by line segments in black. Ice-shelf perimeters are shown as a thin black line. The central circle demarcates the area not surveyed by the satellites (south of 81.5°S). Original data were interpolated for mapping purposes (percentage area surveyed of each ice shelf is provided in table S1). Background is the Landsat Image Mosaic of Antarctica (LIMA).

Fig. 2 Variability in the rate of Antarctic ice-shelf thickness change (meters per year).

Maps for (columns from left to right) Filchner-Ronne, Amundsen, and Ross ice shelves (locations in the bottom right corner) showing average rate of thickness change for (rows) four consecutive 4.5-year intervals (1994–1998.5, 1998.5–2003, 2003–2007.5, and 2007.5–2012). Shorter-term rates can be higher than those from an 18-year interval. Ice-shelf perimeters are thin black lines, and the thick gray line demarcates the limit of satellite observations.

Fig. 3 Time series of cumulative thickness change relative to series mean for Antarctic ice-shelf regions (1994–2012).

Time series correspond to averages for all ice-shelf data within the Antarctic regions defined in Fig. 1. Dots represent average thickness change every 3 months. Error bars are small (in many cases, smaller than the symbols themselves, thus omitted from the plots), making the interannual fluctuation shown by the dots significant. The blue curve is the long-term trend from polynomial regression with the 95% confidence band (18), and the red line shows the regression line to the segment of our data set that overlaps with the period used for a prior ICESat-based analysis (2003–2008) (5). Average rates (in meters per decade) are derived from the end points of the polynomial models.

Ice-shelf average thinning rates from the 18-year polynomial fits in the Amundsen Sea region (AS) range from 1.5 ± 0.9 m/decade for Abbot to 31.1 ± 5.4 m/decade for Crosson, with local maximum thinning of 66.5 ± 9.0 m/decade on Getz (fig. S1 and table S1). Crosson and Getz have lost ~18 and 6% of their thicknesses, respectively, over the 18-year period. If this thinning persists for these two ice shelves, we can expect volume losses of ~100 and 30%, respectively, in the next 100 years. Getz is the single largest contributor to the overall volume loss of Antarctic ice shelves, with an average change of –54 ± 5 km3/year, accounting for ~30% of the total volume loss from the West Antarctic ice shelves (table S1). We find the most dramatic thickness reduction on Venable Ice Shelf in the Bellingshausen Sea (BS), with an average (and maximum) thinning rate of 36.1 ± 4.4 (64.4 ± 4.9) m/decade, respectively (fig. S1 and table S1). This ice shelf has lost 18% of its thickness in 18 years, which implies complete disappearance in 100 years.

For the ice shelves in the AS, observed rates are highest near the deep grounding lines, with lower rates found toward the shallower ice fronts (Fig. 2, table S1, and movie S1). This pattern is consistent with enhanced melting underneath the ice shelf forced by an increased flux of circumpolar deep water (CDW) from across the continental shelf and into the sub–ice-shelf cavity (12, 23, 24). The consequent loss of ice-shelf buttressing from increased ocean-forced melting may have driven the grounding lines inland (25) to a point on a retrograde bed slope at which the marine ice-sheet instability mechanism can take over the dynamics of ice export (7, 26). Hence, observed ice-shelf thinning reflects both ocean-induced basal melting and increased strain rates resulting from faster flows. Our analysis shows that thinning was already under way at a substantial rate at the start of our record in 1994.

On the eastern side of the Antarctic Peninsula [comprising Larsen B (Scar Inlet remnant), Larsen C, and Larsen D], the regional ice-shelf thinning rate of 3.8 ± 1.1 m/decade (Fig. 3) is about half of that on the western side (BS) (Fig. 1). The onset of thinning for Larsen C has progressed southward (Fig. 4), which is consistent with climate-driven forcing discussed in earlier studies (22, 27). The highest thinning rates on Larsen C (with local maximum thinning of 16.6 ± 8.1 m/decade) are near Bawden Ice Rise (Figs. 1 and 4). Assuming that half of this observed thinning is due to air loss within the firn column, and considering that the ice shelf is ~40 m above flotation over the ice rise (28), we can expect Larsen C to fully unground from this pinning point within the next 100 years, with potential consequences on the ice-shelf stability (29).

Fig. 4 Evolution of the rate of thickness change in the Antarctic Peninsula.

Instantaneous rate-of-thickness (meters per year) change for four specific times (1994, 1997, 2000, and 2008) is calculated as the derivative of the polynomial fit to the thickness-change time series. The rate increases spatially with time from north to south in the Larsen Ice Shelf (movie S1). The eastern (Weddell Sea) side of the Antarctic Peninsula (top) shows independent behavior from the western (Bellingshausen Sea) side (bottom).

The regional time-varying trends for the ice shelves in the three East Antarctic regions (Queen Maud, Amery, and Wilkes) are coherent (Fig. 3). Ice shelves in the Wilkes region are challenging for conventional radar altimeters because many of them are small, contained in narrow embayments, and have rough surfaces so that altimeter-derived height changes do not necessarily reflect thickness change accurately. Our estimate of overall thickness change for the Wilkes ice shelves is 1.4 ± 1.5 m/decade, which is not significantly different from zero. The Queen Maud region ice shelves show an overall increase in thickness of 2.0 ± 0.8 m/decade.

Like the AS ice shelves, Totten and Moscow University ice shelves in the Wilkes region buttress a large marine-based section of the East Antarctic ice sheet so that their stability is potentially important to grounded-ice loss. Although these ice shelves were previously reported as thinning (5) on the basis of a straight-line fit to a 5-year record from a satellite laser altimeter (ICESat, 2003–2008), our results show that those estimates are not representative of the longer-term trends (fig. S1B). Our estimate of thickness loss during 2003–2008 is similar to the ICESat-based result, but the full 18-year period shows thickness trends that are not significantly different from zero (fig. S2).

For most ice shelves, our estimates are significantly different from previous results (table S2). Several factors contribute to this. (i) The areas of ice shelves over which measurements are averaged vary between studies, affecting estimates on small ice shelves with large thickness-change signals. (ii) Because of our grid resolution, ice shelf mask, and limited data coverage, we cannot sample near the grounding line of some ice shelves (such as Pine Island or Dotson); in such cases, our estimated changes are likely to represent a lower bound (changes could be larger). (iii) Radar altimeters are less sensitive than are laser altimeters to variations in surface mass balance owing to penetration of the radar signal into the firn layer. (iv) Short records and previous trend-extraction approaches could not capture and account for fluctuations in the underlying trend (fig. S3). This is the dominant factor affecting comparisons between our results and previous studies.

The total volume of East Antarctic ice shelves increased during 1994–2003 by 148 ± 45 km3/year, followed by moderate loss (56 ± 37 km3/year), whereas West Antarctic ice shelves exhibited persistent volume loss over the 18 years, with marked acceleration after 2003 (Fig. 1). Before and after 2003, this region lost volume by 144 ± 45 and 242 ± 47 km3/year, respectively, corresponding to ~70% increase in the average loss rate. The total circum-Antarctic ice-shelf volume loss was negligible (25 ± 64 km3/year) during 1994–2003 and then declined rapidly by 310 ± 74 km3/year after 2003. Overall, from 1994 to 2012 Antarctic ice-shelf volume changed on average by –166 ± 48 km3/year, with mean acceleration of –31 ± 10 km3/year2 (–51 ± 33 km3/year2 for the period 2003–2012).

We have shown that Antarctic ice-shelf volume loss is accelerating. In the Amundsen Sea, some ice shelves buttressing regions of grounded ice that are prone to instability have experienced sustained rapid thinning for almost two decades. If the present climate forcing is sustained, we expect a drastic reduction in volume of the rapidly thinning ice shelves at decadal to century time scales, resulting in grounding-line retreat and potential ice-shelf collapse. Both of these processes further accelerate the loss of buttressing, with consequent increase of grounded-ice discharge and sea-level rise. On smaller scales, ice-shelf thickness variability is complex, demonstrating that results from single satellite missions with typical durations of a few years are insufficient to draw conclusions about the long-term response of ice shelves. Large changes occur over a wide range of time scales, with rapid variations of ice-shelf thickness suggesting that ice shelves can respond quickly to changes in oceanic and atmospheric conditions.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

Table S1 and S2

References (3046)

Movie S1

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Corrections include lag of the satellite’s leading-edge tracker (retracking), surface scattering variations, surface slope, dry atmospheric mass, water vapor, the ionosphere, solid Earth tide, ocean tide and loading, atmospheric pressure, and regional sea-level variation (18).
  3. Acknowledgments: This work was funded by NASA [awards NNX12AN50H 002 (93735A), NNX10AG19G, and NNX13AP60G]. This is ESR contribution 154. We thank J. Zwally’s Ice Altimetry group at the NASA Goddard Space Flight Center for distributing their RA data sets for all satellite radar altimeter missions ( We thank C. Davis and D. Wingham for RA-processing advice. We thank A. Shepherd and anonymous reviewers for their comments on the manuscript.
View Abstract


Navigate This Article