LETTERS
Warming of the Antarctic ice-sheet surface since the
1957 International Geophysical Year
Eric J. Steig
1
, David P. Schneider
2
, Scott D. Rutherford
3
, Michael E. Mann
4
, Josefino C. Comiso
5
& Drew T. Shindell
6
Assessments of Antarctic temperature change have emphasized the
contrast between strong warming of the Antarctic Peninsula and
slight cooling of the Antarctic continental interior in recent
decades
1
. This pattern of temperature change has been attributed
to the increased strength of the circumpolar westerlies, largely in
response to changes in stratospheric ozone
2
. This picture, however,
is substantially incomplete owing to the sparseness and short dura-
tion of the observations. Here we show that significant warming
extends well beyond the Antarctic Peninsula to cover most of West
Antarctica, an area of warming much larger than previously
reported. West Antarctic warming exceeds 0.1 6C per decade over
the past 50 years, and is strongest in winter and spring. Although
this is partly offset by autumn cooling in East Antarctica, the
continent-wide average near-surface temperature trend is positive.
Simulations using a general circulation model reproduce the essen-
tial features of the spatial pattern and the long-term trend, and we
suggest that neither can be attributed directly to increases in the
strength of the westerlies. Instead, regional changes in atmospheric
circulation and associated changes in sea surface temperature and
sea ice are required to explain the enhanced warming in West
Antarctica.
Recent changes in Antarctic ice-sheet surface temperatures appear
enigmatic when compared with global average temperature trends.
Although the Antarctic Peninsula is one of the most rapidly warming
locations on Earth, weather stations on the Antarctic continent generally
show insignificant trends in recent decades
1
. However, all but two of the
continuous records from weather stations are near the coast, providing
little direct information on conditions in the continental interior. The
widely used weather forecast reanalysis data are known to have errors
owing to inconsistent assimilation skill in the satellite and pre-satellite
eras
3
.
In this Letter, we use statistical climate-field-reconstruction tech-
niques to obtain a 50-year-long, spatially complete estimate of
monthly Antarctic temperature anomalies. In essence, we use the
spatial covariance structure of the surface temperature field to guide
interpolation of the sparse but reliable 50-year-long records of 2-m
temperature from occupied weather stations. Although it has been
suggested that such interpolation is unreliable owing to the distances
involved
1
, large spatial scales are not inherently problematic if there is
high spatial coherence, as is the case in continental Antarctica
4
.
Previous reconstructions of Antarctic near-surface temperatures
have yielded inconsistent results, particularly over West Antarctica,
where records are few and discontinuous
5–7
. We improve upon this
earlier work in several ways. We use two independent estimates of the
spatial covariance of temperature across the Antarctic ice sheet: sur-
face temperature measurements from satellite thermal infrared (T
IR
)
1
Department of Earth and Space Sciences and Quaternary Research Center, University of Washington, Seattle, Washington 98195, USA.
2
National Center for Atmospheric Research,
Boulder, Colorado 80307, USA.
3
Department of Environmental Science, Roger Williams University, Bristol, Rhode Island, USA.
4
Department of Meteorology, and Earth and
Environmental Systems Institute, Pennsylvania State University, University Park, Pennsylvania 16802, USA.
5
NASA Laboratory for Hydrospheric and Biospheric Sciences, NASA
Goddard Space Flight Center, Greenbelt, Maryland 20771, USA.
6
NASA Goddard Institute for Space Studies and Center for Climate Systems Research, Columbia University, New York,
New York 10025, USA.
–1 10
r or RE
ab
cd
ef
Figure 1 | Verification and upper-limit calibration statistics calculated for
each grid point from the comparison of reconstructed and original satellite-
derived monthly temperature anomalies. a, Calibration r, 1982–1994.5;
b, calibration r, 1994.5–2006; c, verification r, 1994.5–2006; d, verification r,
1982–1994.5; e, verification RE, 1994.5–2006; f, verification RE,
1982–1994.5. Warm colours in e and f (RE scores greater than zero) show
where results are more accurate than the climatological mean temperature.
Mean grid-point verification results are RE 5 0.11, CE 5 0.09 and r 5 0.46.
Crosses show locations of occupied weather stations.
Vol 457 | 22 January 2009 | doi:10.1038/nature07669
459
Macmillan Publishers Limited. All rights reserved©2009

observations
8
, and up-to-date automatic weather station (AWS) mea-
surements of near-surface air temperature. We use a method
9,10
adapted from the regularized expectation maximization algorithm
11
(RegEM) for estimating missing data points in climate fields. RegEM
is an iterative algorithm similar to principal-component analysis, used
as a data-adaptive optimization of statistical weights for the weather
station data. Unlike simple distance-weighting
5,6
or similar
7
calcula-
tions, application of RegEM takes into account temporal changes in
the spatial covariance pattern, which depend on the relative impor-
tance of differing influences on Antarctic temperature at a given time.
Furthermore, the iterative nature of RegEM allows it to be used with
discontinuous time series, permitting us to take full advantage of the
data available from occupied weather stations. We assess reconstruc-
tion skill using reduction-of-error (RE) and coefficient-of-efficiency
(CE) scores as well as conventional correlation (r) scores. Such veri-
fication metrics are lacking in previous Antarctic temperature
reconstructions
5–7
, but are required for demonstrating skill relative
to the climatological mean and are therefore critical for confidence
in the calculation of temporal trends
10
. Skill metrics for our T
IR
-based
reconstruction from split calibration and verification experiments are
significant (.99% confidence) at all grid points except in some
restricted areas, mostly on the eastern side of the Antarctic
Peninsula (Fig. 1).
Results from our AWS-based reconstruction agree well with those
from the T
IR
data (Fig. 2). This is important because the infrared data
are strictly a measure of clear-sky temperature
8
and because surface
temperature differs from air temperature 2–3 m above the surface, as
measured at occupied stations or at AWSs. Trends in cloudiness or in
the strength of the near-surface inversion could both produce spuri-
ous trends in the temperature reconstruction. The agreement
between the reconstructions, however, rules out either potential bias
as significant. Furthermore, detrending of the T
IR
data before recon-
struction demonstrates that the results do not depend strongly on
trends in said data (Supplementary Information).
Our reconstructions show more significant temperature change in
Antarctica (Fig. 2), and a different pattern for that change than
reported in some previous reconstructions
5,7
(Fig. 3). We find that
West Antarctica warmed between 1957 and 2006 at a rate of
0.17 6 0.06 uC per decade (95% confidence interval). Thus, the area
of warming is much larger than the region of the Antarctic Peninsula.
The peninsula warming averages 0.11 6 0.04 uC per decade. We also
find significant warming in East Antarctica at 0.10 6 0.07 uCper
decade (1957–2006). The continent-wide trend is 0.12 6 0.07 uCper
decade. In the reconstruction based on detrended T
IR
data, warming in
West Antarctica remains significant at greater than 99% confidence,
and the continent-wide mean trend remains at 0.08 uC per decade,
although it is no longer demonstrably different from zero (95% con-
fidence). This is in good agreement with ref. 6, which reported average
continent-wide warming of 0.082 uC per decade (1962–2003) and
shows overall warming in West Antarctica, although statistical signifi-
cance could not be demonstrated owing to the shorter length and
greater variance of the reconstruction. We emphasize that, in general,
1960 1970 1980 1990 2000
−2
−1
0
1
Year
T
e
m
p
e
r
a
t
u
r
e

a
n
o
m
a
l
y

(
°
C
)
b

−1
0
1
2
a


Figure 2 | Reconstructed annual mean Antarctic temperature anomalies,
January 1957 to December 2006. a, East Antarctica; b, West Antarctica.
Solid black lines show results from reconstruction using infrared satellite
data, averaged over all grid points for each region. Dashed lines show the
average of reconstructed AWS data in each region. Straight red lines show
average trends of the T
IR
reconstruction. Verification results for the
continental mean of the T
IR
reconstruction are RE 5 0.34, CE 5 0.31 and
r 5 0.73. Grey shading, 95% confidence limits.
–0.5 0.50
Temperature trend (°C per decade)
b
cd
ef
+0.1
–0.1
+1.1
+0.45
NS
NS
NS
NS
NS
NS
Figure 3 | Spatial pattern of temperature trends (degrees Celsius per
decade) from reconstruction using infrared (T
IR
) satellite data. a, Mean
annual trends for 1957–2006; b, Mean annual trends for 1969–2000, to
facilitate comparison with ref. 2. c–f, Seasonal trends for 1957–2006: winter
(June, July, August; c); spring (September, October, November; d); summer
(December, January, February; e); autumn (March, April, May; f). Black
lines enclose those areas that have statistically significant trends at 95%
confidence (two-tailed t-test). Where it would otherwise be unclear, NS (not
significant) refers to areas of insignificant trends. Red circles and adjacent
numbers in a show the locations of the South Pole and Vostok weather
stations and their respective trends (degrees Celsius per decade) during the
same time interval as the reconstruction (1957–2006). Black circles in b show
the locations of Siple and Byrd Stations, and the adjacent numbers show
their respective trends
13
for 1979–1997.
LETTERS NATURE | Vol 457 | 22 January 2009
460
Macmillan Publishers Limited. All rights reserved©2009