LETTERS

PUBLISHED ONLINE: 11 NOVEMBER 2012 | DOI: 10.1038/NGEO1627

Wind-driven trends in Antarctic sea-ice drift

Paul R. Holland1 * and Ron Kwok2
The sea-ice cover around Antarctica has experienced a slight
expansion in area over the past decades1,2 . This small overall
increase is the sum of much larger opposing trends in different
sectors that have been proposed to result from changes in
atmospheric temperature or wind stress35 , precipitation6,7 ,
ocean temperature8 , and atmosphere or ocean feedbacks9,10 .
However, climate models have failed to reproduce the overall
increase in sea ice11 . Here we present a data set of satellitetracked sea-ice motion for the period of 19922010 that reveals
large and statistically significant trends in Antarctic ice drift,
which, in most sectors, can be linked to local winds. We
quantify dynamic and thermodynamic processes in the internal
ice pack and show that wind-driven changes in ice advection
are the dominant driver of ice-concentration trends around
much of West Antarctica, whereas wind-driven thermodynamic
changes dominate elsewhere. The ice-drift trends also imply
large changes in the surface stress that drives the Antarctic
ocean gyres, and in the fluxes of heat and salt responsible for
the production of Antarctic bottom and intermediate waters.
Satellite studies of Antarctic sea-ice motion1214 have greatly
advanced our understanding of its large-scale dynamics, but their
short length and combination of different sensors have precluded
an investigation of decadal ice-motion trends. We present a new
consistent data set for AprilOctober 19922010 that is specifically
designed to address this question (see Methods). The long-term
mean of these ice-motion data is in good agreement with shorter
data sets reported previously1214 . Away from coastlines, the ice is
close to a state of free drift, moving at a small (spatially variable)
angle to the left of the geostrophic wind14 , which follows contours
of atmospheric pressure (Fig. 1a). This creates a tendency for the ice
to circulate around the three climatological lows in the circumpolar
pressure trough that surrounds Antarctica, leading to maximum
ice export in the Weddell, Cooperation and Ross seas. A nearly
continuous westward current dominates ice flow next to the coast.
The freely drifting areas have a strong correlation between ice
motion and winds14 . Figure 1b shows a map of the vector correlation between interannual time series of AprilOctober mean ice
motion and 10-m winds from the ERA-Interim atmospheric reanalysis (see Methods). The spatial mean of the correlations is r 2 = 0.52,
but large areas in the Pacific and Atlantic sectors have r 2 > 0.7. This
quantifies the effect of winds on ice motion through both direct
wind stress and any portion of the ocean stress and sea-surface tilt
forcing that is correlated to local winds. The remaining ice-motion
variance can be attributed to error in the motion-tracking and
atmosphere reanalysis model, ocean forcing decorrelated from local
winds, and modification of ice motion by internal stresses. Lowcorrelation areas in Fig. 1b are generally associated with convergent
ice motion or flow near coastlines, where internal stresses are large.
Our primary advance is the discovery of large and spatially
variable decadal trends in ice motion (Fig. 2a), with statistically
significant ice-speed changes (trend multiplied by period) of up to
30% occurring over 19 years. Changes in the meridional component

of ice motion, which is largely responsible for the size and shape
of the Antarctic ice cover, are even greater. Widespread increases
of 2 cm s1 in northward export in the Ross and Amundsen seas
equal the mean northward drift during this period (2 cm s1 ),
and similar northward changes (2 cm s1 ) occur in the Kong
Hkon VII Hav (mean 4 cm s1 ; King Hkon Sea hereafter). In
contrast, northward ice export in the Weddell Sea decreased by
1 cm s1 (mean 4 cm s1 ), focused on a strong and significant
decrease in export from the Ronne Polynya in the southwest.
Bellingshausen Sea ice experienced a large but insignificant change
of 3 cm s1 southward, implying great variability around its
mean of 1 cm s1 northward. Opposing ice-drift trends in the
Ross and Bellingshausen seas have been inferred previously from
atmospheric model winds4,15 , but are observed and quantified here
for the first time. Changed ice motion in the southernmost Ross and
Weddell seas has been shown to affect dense-water formation over
limited regions16,17 but our data reveal widespread and coherent
drift anomalies north of the near-shore polynyas.
The strong correlation between observed ice motion and
reanalysis winds in most of the sea-ice zone implies that ice-motion
trends are largely caused by wind trends (Fig. 2b). In fact, the
correlations in Fig. 1b specifically quantify this linkage because they
are calculated between the data sets used to calculate the trends
shown in Fig. 2 (see Methods). Modelled surface pressure trends
suggest that opposing changes in the Ross and Bellingshausen seas
are responding to a large but statistically insignificant reduction
in pressure centred on the Amundsen Sea, whereas changes in
the Weddell and King Hkon seas are caused by a smaller but
significant increase in pressure over the prime meridian. The high
correlations cross-validate the ice motion tracking and reanalysis
model in these regions. Agreement between the trends is poorer
around East Antarctica, as expected given the proximity of the ice to
the coast in this region. However, some disagreements, such as that
in the Mawson Sea, are large enough to imply that the ice-motion
data or reanalysis winds may be erroneous.
The ice-motion trends provide new insight into the well-known
but poorly understood changes in Antarctic ice concentration.
Concentration trends are largest in autumn1,2,4,9 , and geographical
trend patterns persist largely unaltered until spring18 . This suggests
that most winter anomalies emanate from large changes in the
autumn ice-edge advance, possibly as a result of summertime
iceocean feedbacks9,19 . Autumn concentration trends for 1992
2010 (Fig. 3a) mimic those described previously for the full post1978 satellite record4,18 , with large losses at the ice edge in the
Bellingshausen and Mawson (2% yr1 ) and Weddell (1% yr1 )
seas outweighed by widespread gains in the Ross (1% yr1 )
and King Hkon and Cosmonaut (0.5% yr1 ) seas. Overlaying
ice-motion trend vectors onto these changes suggests a link, with
regions of significantly increased northward ice flow exhibiting
significant increases in ice concentration and vice versa. This link
seems strongest in the Atlantic and Pacific sectors, notably failing in
the Mawson Sea. Figure 3b shows a strong and significant autumn

1 British

Antarctic Survey, Cambridge CB3 0ET, UK, 2 Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California 91109, USA.
*e-mail: p.holland@bas.ac.uk.
872

NATURE GEOSCIENCE | VOL 5 | DECEMBER 2012 | www.nature.com/naturegeoscience

2012 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO1627

LETTERS
a

a
1,0
15

99

05

1,0
5

990

0.07

984 hPa

0.0 m s1

978

0.07

0.2 m s1

0.005 m s1 yr1

tic

Atlan

KS
CTS
WS
1.0

0.6
CNS

0.5

0.0 hPa yr1

Indian

BS

0.0

0.6
0.1 m s1 yr1

MS

AS

RS
DS
ic

Pa
cif

Figure 1 | AprilOctober 19922010 mean ice motion and its relation to

wind forcing. Wind and ice motion are strongly coupled in Pacific and
Atlantic sectors but weakly related around East Antarctica, where a coastal
current of ice dominates. a, Mean ice-motion vectors overlaid on
ERA-Interim reanalysis sea-level pressure. b, Vector correlation between
ice-motion vectors and ERA-Interim 10-m wind vectors (white contours
show r2 = 0.4, significant at 99%). WS, Weddell Sea; KS, King Hkon Sea;
CTS, Cosmonaut Sea; CNS, Cooperation Sea; MS, Mawson Sea;
DS, Dumont DUrville Sea; RS, Ross Sea; AS, Amundsen Sea;
BS, Bellingshausen Sea.

deepening over the Amundsen Sea, thought to result from the

increased intensity of the Southern Annular Mode4,5,15 . Autumn icemotion trends are again closely related to wind trends in the Atlantic
and Pacific sectors and disagree elsewhere. A significant trend
towards northerly winds occurs in the Mawson Sea, suggesting
a wind-driven decrease in ice concentration that is not reflected
in ice-motion trends.
The general link between trends in ice concentration and ice
motion or winds seems strong. However, this cannot separate
dynamic from thermodynamic causes, because more southerly
winds could increase the ice concentration through either increased

Figure 2 | AprilOctober 19922010 ice-motion trends and their relation

to wind forcing. Large and statistically significant changes in ice motion are
driven by changes in the winds. a, Ice-motion trend vectors overlaid on
19-year change in meridional ice speed (change is linear trend multiplied by
period, positive northwards; black vectors have meridional ice-motion
trends significant at >90%). b, ERA-Interim reanalysis 10-m wind trend
vectors overlaid on trend in sea-level pressure (white and grey contours
show pressure trends significant at 90% and 95%; black vectors have
meridional wind trends significant at >90%; magenta contour shows
extent of motion trends).

transport of ice from the south or through an atmospheric cooling

from increased advection of cold polar air masses35 . Similarly,
more northerly winds would decrease ice transport and cause
atmospheric warming. Our new data present a unique opportunity
to quantify these processes by decomposing the ice-concentration
budget in the internal ice pack into dynamic and thermodynamic
contributions. The ice concentration change and its dynamic
contribution are calculated by combining concentration and
motion data, and their residual is found to represent the
thermodynamic contribution (see Supplementary Methods).

NATURE GEOSCIENCE | VOL 5 | DECEMBER 2012 | www.nature.com/naturegeoscience

2012 Macmillan Publishers Limited. All rights reserved.

873

NATURE GEOSCIENCE DOI: 10.1038/NGEO1627

LETTERS
a

0.02
0.0 yr1
0.02
0.005 m s1 yr1

0.6
0.0 hPa yr1
0.6
0.1 m s1 yr1

Figure 3 | Autumn (AprilJune) 19922010 ice motion and concentration

trends and their relation to wind forcing. Wind-driven changes in ice
motion are clearly linked to changes in ice concentration. a, Ice-motion
trend vectors overlaid on ice-concentration trends. b, ERA-Interim 10-m
wind trend vectors overlaid on trend in sea-level pressure. White, grey and
black contours show underlay field trends significant at 90%, 95% and
99% respectively; black vectors have meridional trends significant at
>90%; magenta contour in b shows extent of concentration trends.

Before examining trends, it is instructive to first decompose the

mean AprilOctober ice-concentration budget20 (Supplementary
Fig. S1). Freezing in the inner pack is maintained by divergence,
which supports greater oceanatmosphere heat exchange than is
otherwise possible in consolidated ice. In the main export regions,
ice is advected to the margins and then melted, because it is
thermodynamically unsustainable there even in winter; the ice
cover in these regions extends hundreds of kilometres further
equatorward than it would in the absence of northward advection.
This illustrates the Antarctic sea-ice freshwater pump, which
contributes brine to Antarctic Bottom Water close to the continent
and fresh water to Antarctic Intermediate Water and mode
874

waters in the ice-melting zone. In areas of mean northerly winds

(Bellingshausen, Cosmonaut and Dumont DUrville seas; Fig. 1a)
southward advection opposes thermodynamic growth of the ice
cover, and freezing extends closer to the ice edge.
The mean concentration difference over autumn is dominated
by freezing, with advection and divergence being minor contributors during this period (Supplementary Fig. S2). However, in
the Pacific sector and Weddell Sea, trends in the autumn concentration difference seem to be strongly influenced by dynamics
(Supplementary Fig. S3). In contrast, trends in the King Hkon
Sea are controlled by thermodynamics. Supplementary Fig. S4
shows the proportion of the autumn concentration difference
trend that is explained by trends in dynamical processes. The ratio
is noisy, but after heavy smoothing, it confirms that dynamic
trends dominate in the Pacific sector. Trends in freezing in this
sector can actually oppose the ice-concentration changes, because
dynamical processes are progressively replacing thermodynamics.
Ice-concentration losses in the Weddell Sea also seem to be caused
by decreased northward advection, but the concentration increase
in King Hkon Sea and other changes around East Antarctica
contain a strong thermodynamic component. The wind trends in
these regions suggest that changes in cold- and warm-air advection
explain the thermodynamic trends.
The ultimate cause of the wind and ice changes lies in the largescale climate variability of the Southern Hemisphere. Antarctic sea
ice can contain 35-year cyclic anomalies that might be partly
aliased into our calculations1,16,21 , but our trends cover several
such cycles and are consistent with longer-term studies18 . Aspects
of the wind trends (and therefore ice-motion trends) can be
attributed to large-scale modes such as the Southern Annular
Mode and El Nio/Southern Oscillation3,19,22 . Modern trends in
these modes could arise through natural variability, but some
evidence suggests that they are forced by the Southern Hemisphere
ozone hole and increased greenhouse gases4,23 . Our conclusions
that ice-motion trends are dominated by winds, and that winds
contribute significantly to ice concentration trends through both
dynamic and thermodynamic effects, reinforce the need for a better
understanding of both the wind changes and the anthropogenic
forcing of relevant climate modes.
Our conclusions are of fundamental importance in rectifying
the failure of present climate models to hindcast the recent
increase in Antarctic sea ice11 . In particular, they suggest that
surface winds and ice dynamics and thermodynamics must be
accurately represented. Our data set provides an observational map
of changes against which models can be compared, and any faults
can be diagnosed using our decomposition of the ice-concentration
budget into dynamic and thermodynamic components. When
climate models can hindcast ice-concentration increases we will
have good reason to believe their forecasted ice loss under the
effects of climate change.
Our data offer a new view of surface change relevant to all
components of the Antarctic climate. The good fit between ice
motion and reanalysis wind trends, in an area of extremely sparse
in situ data, is testament to the power of satellite sounder data
assimilation into ERA-Interim. It also implies confidence that these
winds can be used to force models of Antarctic ice and ocean
trends over recent decades. The large and widespread changes in
ice motion imply considerable changes in sea-surface forcing, both
directly through observed changes in ice stress and indirectly by
validating trends in reanalysis wind stress. The ice-motion trends
suggest that increased cyclonic forcing has accelerated the Ross
Gyre, supporting its possible involvement in ice-sheet melting and
Ross Sea freshening24 . The decrease in Weddell Sea ice cyclonicity
suggests a Weddell Gyre deceleration, implying that gyre changes
alone cannot explain the warming of Antarctic Bottom Water
exported to the abyssal Atlantic25 . Changes in meridional ice
NATURE GEOSCIENCE | VOL 5 | DECEMBER 2012 | www.nature.com/naturegeoscience

2012 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO1627

export also affect freshwater budgets, implying greater brine release
in the southern Ross16,17 and King Hkon seas and increased
meltwater input to their north, which may have contributed to the
freshening of Antarctic Intermediate Water26 . The data presented
here thus offer a new observational insight into many facets of the
Antarctic climate system.

Methods
We produced a data set of over 5 million individual ice-motion vectors
on a 100-km polar stereographic grid by tracking brightness-temperature
patterns in successive daily images from Special Sensing Microwave/Imager
(SSMI) instruments16,27 . Specifically, we used the 85 GHz vertically polarized
channels of the SSMI instruments aboard satellites F8, F11, F13 and F17 of the
Defense Meteorological Satellites Program. Supporting analyses used NASA
(National Aeronautics and Space Administration) Team SSMI-derived ice
concentrations28 (http://nsidc.org/data/nsidc-0051.html) and the ERA-Interim
atmospheric reanalysis (http://data-portal.ecmwf.int/data/d/interim_daily/) for
the same period. Differences between the ice-motion data and analyses from
higher-resolution satellite data and in situ buoys have mean 00.5 cm s1 and
standard deviation 48 cm s1 (refs 27,29), levels of accuracy consistent with similar
data sets13 . SSMI ice-concentration data are accurate to approximately 5% for the
winter measurements presented here28 .
We restrict our analyses to only the period of continuous daily sampled 85 GHz
data, 19922010, producing a consistently sampled and high-quality record that
allows the unambiguous identification of long-term trends. The motion-tracking
procedure is less accurate when the ice is experiencing surface melt, causing more
ice-motion data to be rejected during summer months, so to ensure a consistent
coverage we use only data from AprilOctober. Data are also often missing near
the ice edge, because fast day-to-day changes in ice features are not resolvable
by the tracking procedure. These seasonal and spatial limitations imply that our
conclusions are limited to the behaviour of only the internal ice pack, away from the
ice edge, and only in non-summer months. This is an important limitation because
recent studies suggest that the Antarctic trends are related to modulation of the
advance and retreat of the ice edge by summertime interaction with the ocean9 .
Interannual trends at each grid point are calculated from yearly averages that
are taken over either the whole observation period in each year (AprilOctober) or
only from autumn data, defined here as AprilJune to maximize the use of our data.
A fraction of the annual ice advance (and its trends) occurs earlier in the year in
some regions9 , for which we have insufficient data. Trends are calculated separately
for each component of a vector, and these vector trends are then plotted. The significance of a vector trend is itself a vector, which complicates its illustration in figures.
We choose to highlight the significance of trends only in the meridional direction,
because trends in northward ice export generally have the most important impacts.
The field of vector correlations presented in Fig. 1b is designed to quantify
the linkage between interannual trends in winds and ice motion, so AprilOctober
daily vectors in each quantity are first averaged into yearly means, and then the
19-year time series of ice motion and wind vectors are correlated at each grid point.
The relationship between time series of wind and ice-drift vectors is examined
using a vector correlation measure that can be viewed as a generalization of the
simple correlation coefficient between two scalar time series30 . According to the
method, the two vector time series at each grid point would be perfectly correlated
if the magnitude of the vectors were related by a constant factor and their direction
subject to a constant offset. The resulting r 2 is scaled to lie between 0 and 1 here
for ease of comparison with scalar correlation values. If the two vector series are
independent, nr 2 is asymptotically distributed as 2 with four degrees of freedom,
but for smaller samples (n  64) its distribution can be calculated empirically30 .
This can be used to assess the significance of correlations; for n = 19 and using r 2
scaled to lie between 0 and 1 we estimate that a correlation of r 2 = 0.4 provides a
significance level of 99% (ref. 30).

Received 13 April 2012; accepted 9 October 2012; published online

11 November 2012

References
1. Zwally, H. J., Comiso, J. C., Parkinson, C. L., Cavalieri, D. J. & Gloersen, P.
Variability of Antarctic sea ice 19791998. J. Geophys. Res. 107, 3041 (2002).
2. Comiso, J. C. & Nishio, F. Trends in the sea ice cover using enhanced
and compatible AMSR-E, SSM/I, and SMMR data. J. Geophys. Res. 113,
C02s07 (2008).
3. Liu, J. P., Curry, J. A. & Martinson, D. G. Interpretation of recent Antarctic sea
ice variability. Geophys. Res. Lett. 31, L02205 (2004).
4. Turner, J. et al. Non-annular atmospheric circulation change induced by
stratospheric ozone depletion and its role in the recent increase of Antarctic sea
ice extent. Geophys. Res. Lett. 36, L08502 (2009).
5. Lefebvre, W. & Goosse, H. Influence of the Southern Annular Mode on the sea
ice-ocean system: The role of the thermal and mechanical forcing. Ocean Sci. 1,
145157 (2005).

LETTERS
6. Liu, J. P. & Curry, J. A. Accelerated warming of the Southern Ocean and its
impacts on the hydrological cycle and sea ice. Proc. Natl Acad. Sci. USA 107,
1498714992 (2010).
7. Manabe, S., Spelman, M. J. & Stouffer, R. J. Transient responses of a coupled
ocean atmosphere model to gradual changes of atmospheric CO2 . Part II:
Seasonal response. J. Clim. 5, 105126 (1992).
8. Jacobs, S. S. & Comiso, J. C. Climate variability in the Amundsen and
Bellingshausen Seas. J. Clim. 10, 697709 (1997).
9. Stammerjohn, S. E., Massom, R., Rind, D. & Martinson, D. G. Regions of rapid
sea ice change: An inter-hemispheric seasonal comparison. Geophys. Res. Lett.
39, L06501 (2012).
10. Zhang, J. L. Increasing Antarctic sea ice under warming atmospheric and
oceanic conditions. J. Clim. 20, 25152529 (2007).
11. Eisenman, I., Schneider, T., Battisti, D. S. & Bitz, C. M. Consistent changes
in the sea ice seasonal cycle in response to global warming. J. Clim. 24,
53255335 (2011).
12. Emery, W. J., Fowler, C. W. & Maslanik, J. A. Satellite-derived maps of
Arctic and Antarctic sea ice motion: 19881994. Geophys. Res. Lett. 24,
897900 (1997).
13. Heil, P., Fowler, C. W. & Lake, S. E. Antarctic sea-ice velocity as derived from
SSM/I imagery. Ann. Glaciol. Ser. 44, 361366 (2006).
14. Kimura, N. Sea ice motion in response to surface wind and ocean current in
the Southern Ocean. J. Meteorol. Soc. Jpn 82, 12231231 (2004).
15. Lefebvre, W., Goosse, H., Timmermann, R. & Fichefet, T. Influence of the
Southern Annular Mode on the sea ice-ocean system. J. Geophys. Res. 109,
C09005 (2004).
16. Comiso, J. C., Kwok, R., Martin, S. & Gordon, A. L. Variability and trends
in sea ice extent and ice production in the Ross Sea. J. Geophys. Res. 116,
C04021 (2011).
17. Drucker, R., Martin, S. & Kwok, R. Sea ice production and export from coastal
polynyas in the Weddell and Ross Seas. Geophys. Res. Lett. 38, L17502 (2011).
18. Comiso, J. C. Polar Oceans from Space (Springer, 2010).
19. Stammerjohn, S. E., Martinson, D. G., Smith, R. C., Yuan, X. & Rind, D.
Trends in Antarctic annual sea ice retreat and advance and their relation
to El Nio-Southern Oscillation and Southern Annular Mode variability.
J. Geophys. Res. 113, C03s90 (2008).
20. Kimura, N. & Wakatsuchi, M. Large-scale processes governing the seasonal
variability of the Antarctic sea ice. Tellus A 63, 828840 (2011).
21. White, W. B. & Peterson, R. G. An Antarctic circumpolar wave in surface
pressure, wind, temperature and sea-ice extent. Nature 380, 699702 (1996).
22. Yuan, X. J. & Li, C. H. Climate modes in southern high latitudes and their
impacts on Antarctic sea ice. J. Geophys. Res. 113, C06s91 (2008).
23. Thompson, D. W. J. et al. Signatures of the Antarctic ozone hole in Southern
Hemisphere surface climate change. Nature Geosci. 4, 741749 (2011).
24. Jacobs, S. S., Giulivi, C. F. & Mele, P. A. Freshening of the Ross Sea during the
late 20th century. Science 297, 386389 (2002).
25. Meredith, M. P. et al. Synchronous intensification and warming of Antarctic
Bottom Water outflow from the Weddell Gyre. Geophys. Res. Lett. 38,
L03603 (2011).
26. Wong, A. P. S., Bindoff, N. L. & Church, J. A. Large-scale freshening
of intermediate waters in the Pacific and Indian oceans. Nature 400,
440443 (1999).
27. Kwok, R., Schweiger, A., Rothrock, D. A., Pang, S. & Kottmeier, C. Sea ice
motion from satellite passive microwave imagery assessed with ERS SAR and
buoy motions. J. Geophys. Res. 103, 81918214 (1998).
28. Cavalieri, D. J., Parkinson, C. L., Gloersen, P & Zwally, H. J. Sea Ice
Concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS Passive
Microwave Data, 19922010 (National Snow and Ice Data Center, 1996).
29. Kwok, R. Ross Sea ice motion, area flux, and deformation. J. Clim. 18,
37593776 (2005).
30. Crosby, D. S., Breaker, L. C. & Gemmill, W. H. A proposed definition for vector
correlation in geophysicstheory and application. J. Atmos. Ocean. Technol.
10, 355367 (1993).

Acknowledgements
Part of this work was performed at the Jet Propulsion Laboratory, California Institute of
Technology, under contract with the National Aeronautics and Space Administration.

Author contributions
P.R.H. designed and performed the research and wrote the manuscript. R.K. provided
the data and contributed to the manuscript.

Additional information
Supplementary information is available in the online version of the paper. Reprints and
permissions information is available online at www.nature.com/reprints. Correspondence
and requests for materials should be addressed to P.R.H.

Competing financial interests

The authors declare no competing financial interests.

NATURE GEOSCIENCE | VOL 5 | DECEMBER 2012 | www.nature.com/naturegeoscience

2012 Macmillan Publishers Limited. All rights reserved.

875