ARTICLES
PUBLISHED ONLINE: 23 DECEMBER 2012 | DOI: 10.1038/NGEO1671
Central West Antarctica among the most rapidly
warming regions on Earth
David H. Bromwich
1
*
†
, Julien P. Nicolas
1†
, Andrew J. Monaghan
2
, Matthew A. Lazzara
3
,
Linda M. Keller
4
, George A. Weidner
4
and Aaron B. Wilson
1
There is clear evidence that the West Antarctic Ice Sheet is contributing to sea-level rise. In contrast, West Antarctic
temperature changes in recent decades remain uncertain. West Antarctica has probably warmed since the 1950s, but there
is disagreement regarding the magnitude, seasonality and spatial extent of this warming. This is primarily because long-term
near-surface temperature observations are restricted to Byrd Station in central West Antarctica, a data set with substantial
gaps. Here, we present a complete temperature record for Byrd Station, in which observations have been corrected, and
gaps have been filled using global reanalysis data and spatial interpolation. The record reveals a linear increase in annual
temperature between 1958 and 2010 by 2.4±1.2
◦
C, establishing central West Antarctica as one of the fastest-warming regions
globally. We confirm previous reports of West Antarctic warming, in annual average and in austral spring and winter, but find
substantially larger temperature increases. In contrast to previous studies, we report statistically significant warming during
austral summer, particularly in December–January, the peak of the melting season. A continued rise in summer temperatures
could lead to more frequent and extensive episodes of surface melting of the West Antarctic Ice Sheet. These results argue for
a robust long-term meteorological observation network in the region.
G
lacier acceleration along the Amundsen Sea coast
1
has been
responsible for the increasing mass loss from the West
Antarctic Ice Sheet (WAIS) in recent years
2
. This has raised
concerns about the present and future state of the WAIS, given its
known potential instability in a warmer climate
3
. Key mechanisms
behind this acceleration have been identified as the melting and
thinning of the floating ice shelves triggered by warm ocean water
4,5
.
In comparison, it is still a matter of debate whether the atmosphere
above the WAIS has warmed over the past few decades, especially
since the 1957–1958 International Geophysical Year, the start of
the instrumental period in Antarctica
6–8
. Unlike Greenland, where
the extent of surface melting has grown markedly
9
, West Antarctica
has not shown any unequivocal signs of atmospheric warming
10,11
.
The question is, therefore, whether West Antarctic temperatures
have, indeed, not significantly changed (or even decreased) since the
1950s; or whether they have increased but not so much as to reach
the melting point at the surface. In other words, could the WAIS be
on the verge of becoming like Greenland? If so, is the exceptionally
warm summer month of January 2005, when widespread surface
melting occurred over a large portion of the WAIS (ref. 12;
Supplementary Fig. S1), an early manifestation of this transition?
Assessing Antarctic climate change on timescales of a few decades
is a well-recognized challenge owing to the paucity of surface
observations. Accordingly, statistical methods have been used to
reconstruct Antarctic near-surface temperatures by interpolating
the sparse meteorological records available since the International
Geophysical Year
6,7,13,14
. These reconstructions have produced
contrasting, and sometimes contradictory, temperature trends over
West Antarctica. This is not surprising as, in this region, the
1
Polar Meteorology Group, Byrd Polar Research Center, and Atmospheric Sciences Program, Department of Geography, The Ohio State University,
Columbus, Ohio 43210, USA,
2
National Center for Atmospheric Research, Boulder, Colorado 80307, USA,
3
Antarctic Meteorological Research Center,
Space Science and Engineering Center, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA,
4
Department of Atmospheric and Oceanic
Sciences, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA.
†
These authors contributed equally to this work.
*e-mail: bromwich.1@osu.edu.
reconstructions can rely only on incomplete observations from a
single site: Byrd Station (80
◦
S, 120
◦
W; Fig. 1). Furthermore, West
Antarctica is climatologically distinct from the rest of the continent,
especially with greater influence from the tropics
15–17
, so that its
climate variability and trends are not necessarily well reflected in
peripheral temperature records.
Here, we present a new reconstruction of the Byrd temperature
record that aims to improve on previous infilling methods
14,18–20
.
The observations from the initial year-round occupied station
(1957–1975) are combined with updated and corrected data from
the automatic weather station (AWS) maintained since 1980 by the
US Antarctic AWS Program
21
. Missing observations are estimated
using adjusted temperature data from the ERA-Interim reanalysis
22
for 1979–2012 and, before 1979, a combination of global reanalysis
data and spatially interpolated observations from other Antarctic
sites. Full details on the corrections to the observations and the
infilling technique are provided in the Methods and Supplementary
Methods. Although a full spatial interpolation of West Antarctic
temperatures lies beyond the scope of this paper, the Byrd record
is expected to provide insight into temperature changes over a large
portion of the WAIS owing to its broad spatial footprint (Fig. 1) .
Improved temperature estimates
The reliability of the temperature observations is a central premise
of our reconstruction. The temperature readings were collected
by professional weather observers until the 1970s, providing a
robust anchor for the early portion of record. The operation of
the AWS has proved more challenging in the harsh Antarctic
environment, but the origins of the data gaps are well understood
21
.
NATURE GEOSCIENCE | VOL 6 | FEBRUARY 2013 | www.nature.com/naturegeoscience 139
© 2013 Macmillan Publishers Limited. All rights reserved.
ARTICLES
NATURE GEOSCIENCE DOI: 10.1038/NGEO1671
Ross Sea
Marie Byrd Land
Amundsen
Sea
Bellingshausen
Sea
Weddell
Sea
Ross
Ice Shelf
Ronne
Ice Shelf
150° W
120° W
120° E
150° E
60° S
70
°
S
80
°
S
30° W
30° E
60° E
FV
Antarctic Peninsula
East
Antarctica
West
Antarctica
Annual temperature correlation with Byrd
¬0.3 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Figure 1 | Map of Antarctica and annual spatial footprint of the Byrd
temperature record. The colour shadings show the correlation between the
annual mean temperatures at Byrd and the annual mean temperatures at
every other grid point in Antarctica. The correlations are computed using
ERA-Interim 2-metre temperature time series from 1979 to 2011. The star
symbol denotes the location of Byrd Station/AWS. The filled black circles
denote the locations of permanent research stations with long-term
temperature records (FV, Faraday/Vernadsky).
A reassessment of the calibration requirements of the AWS
hardware was conducted in 2011 and led to the release of a
corrected set of temperature observations in December 2011 (see
Supplementary Methods).
Global atmospheric reanalyses are, in principle, uniquely suited
for our reconstruction. By synthesizing a wide range of historical
observations with the fields from a state-of-the-art atmospheric
model, these reanalyses provide a best possible representation of
the state of the atmosphere, complete both in space and time.
Yet, the quality of their temperature estimates has long remained
inadequate in Antarctica
19,23,24
, prompting the use of alternative
reconstruction methods
14,18–20
. Compared with other reanalyses,
ERA-Interim predicts the near-surface temperature at Byrd with
markedly greater skill—even without the constraint of Byrd AWS
observations—and could therefore be used for the infilling (see
Methods and Supplementary Figs S8 and S9). The uncertainties of
the reconstruction are larger during 1970–1978 owing to the lower
reliability of the reanalysis data sets available for this period and
the almost complete absence of Byrd observations. This limited
portion of the record is found to have little impact on the long-term
temperature trends (Supplementary Table S1).
Temperature trends at Byrd Station
The reconstructed Byrd record is presented in Fig. 2 as annual
and seasonal mean temperature time series from 1957 to 2011.
For the annual mean temperature, the linear trend calculated for
1958–2010 reveals an average warming of 0.47±0.23
◦
C per decade,
statistically significant at the 99% confidence level (CL; Fig. 3a),
which translates into a temperature increase of 2.44 ±1.19
◦
C in 52
years. This warming is close to that measured at Faraday/Vernadsky,
on the western coast of the Antarctic Peninsula (0.58 ± 0.31
◦
C per
decade), a site already known for its rapid atmospheric warming
25
.
The temperature trends at Byrd and Faraday are substantially
greater than the global average (0.13 ± 0.03
◦
C per decade; ref. 26)
and comparable in magnitude to the warming observed over land
in the Northern Hemisphere high latitudes (Fig. 3c).
Seasonally, warming has occurred at Byrd throughout the year,
yet without statistical significance in austral autumn (March, April,
May (MAM); Fig. 3a). Spring (September, October, November
(SON)) exhibits the largest temperature trend (0.82 ± 0.40
◦
C
per decade) and highest significance level (99%). The trends in
winter (June, July, August (JJA)) and summer (December, January,
February (DJF)) are both statistically significant at the 95% CL
with 0.54 ± 0.51
◦
C per decade and 0.30 ± 0.27
◦
C per decade,
respectively. Importantly, the summertime warming is maximum
and most significant (CL > 99%) in December–January, the two
climatologically warmest months of the year at Byrd and the peak
of the melting season in Antarctica.
Signs of interdecadal variability are also evident in Byrd
temperatures. On the annual scale, most of the warming at
Byrd seems to have occurred during the mid- to late 1980s,
with temperatures apparently levelling off since the early 1990s
(Fig. 2a). Both as a result of slower temperature increases and large
interannual variability (particularly marked in SON in the latter
part of the record), none of the trends attains statistical significance
during 1980–2010, except for December–January (Fig. 3b).
Comparison with other temperature reconstructions
The temperature trends estimated from our Byrd record are con-
trasted with those derived from four sets of Antarctic temperature
reconstructions
6,7,13,14
for 1958–2005 (Fig. 4) and 1958–2001 (Sup-
plementary Table S2). Our record shows (almost systematically)
stronger warming than all other data sets, although, because of the
error bars, the various trend estimates are not always statistically
distinguishable from one another.
There is overall agreement among the reconstructions on
greatest seasonal warming occurring in SON (statistically significant
only in our record and refs 6,14), which corroborates the
conclusions from a recent investigation of the West Antarctic
warming during this season
8
. However, our results and ref. 14
indicate temperature trends at least twice as large as the other
reconstructions. The same two data sets also stand out in DJF
with greater and statistically significant positive trends (marginally
significant in ref. 14). This summer warming is notably at odds
with the marked tropospheric cooling seen in Microwave Sounding
Unit observations available since 1979 (ref. 27). In MAM and
JJA, our reconstructed record agrees relatively well with ref. 6 for
the 1958–2005 period. Both find JJA the second-fastest-warming
season (yet without statistical significance), which confirms other
evidence of surface and tropospheric winter warming over West
Antarctica
17,27,28
; and both show very similar temperature trends
in MAM (significant only in ref. 6). In these two seasons (MAM
and JJA), the other reconstructions
7,13,14
have substantially smaller
trends (or even negative values in ref. 7).
Annually, a pronounced warming in West Antarctica in recent
decades has also been detected in recent borehole temperature
measurements
29,30
, especially at the WAIS Divide drilling site,
160 km northeast of Byrd. The WAIS Divide record
30
, in partic-
ular, suggests a warming occurring later than at Byrd (early to
mid-1990s) and continuing into the 2000s, instead of flattening
out. Although this may reveal greater spatial heterogeneity in the
temperature changes than inferred from the Byrd temperature foot-
print (Fig. 1), the discrepancy may also be related to the nonlinear
temporal smoothing inherent to borehole temperature retrievals.
Investigation of the winter and spring warming
The causes of the West Antarctic warming in JJA and SON have
been investigated in two recent studies
8,17
that have highlighted,
140 NATURE GEOSCIENCE | VOL 6 | FEBRUARY 2013 | www.nature.com/naturegeoscience
© 2013 Macmillan Publishers Limited. All rights reserved.
NATURE GEOSCIENCE DOI: 10.1038/NGEO1671
ARTICLES
¬31
¬30
0
12
Number of observations
Number of observations Number of observations
Number of observations Number of observations
¬29
¬28
¬27
Temperature (°C)
¬26
¬25
¬24
Annual
1960 1970 1980 1990 2000 2010
1960 1970 1980
Year
Year Year
Year Year
1990 2000 2010
1960 1970 1980 1990 2000 2010
1960 1970 1980
Autumn (MAM)
1990 2000 2010
1960 1970 1980 1990 2000 2010
a
b
d
c
¬20
¬36
¬34
¬30
¬32
¬28
¬26
¬24
3
00
3
0
3
0
3
¬19
¬18
¬17
¬16
Temperature (°C)Temperature (°C)
e
Temperature (°C)
Temperature (°C)
¬15
¬14
¬13
Summer (DJF)
Winter (JJA) Spring (SON)
¬44
¬34
¬32
¬30
¬28
¬26
¬24
¬22
¬42
¬40
¬38
¬36
¬34
¬32
¬30
¬28
Figure 2 | Temperature time series from the reconstructed Byrd record. a–e, Annual (a) and seasonal (b–e) mean temperature time series. Red markers
denote the portions of the record for which > 1/3 of the observations are missing; black markers are used otherwise. The solid grey line represents the
centred 5-year moving average temperature. The histograms (right vertical axis) show the number of monthly mean temperature observations available
per year or per season. For summer (DJF), the year refers to January.
in particular, its linkage to lower-latitude sea surface temperature
(SST) changes. Is the temperature variability observed at Byrd
consistent with these findings? And can they also explain the
warming in austral summer?
In JJA, the warming has been associated with an increase in
geopotential heights over West Antarctica
17
(Fig. 5a). This pattern
has promoted onshore winds (warm advection) to Marie Byrd Land
and is consistent with the spatial correlations between Byrd tem-
perature and the 500 hPa geopotential height (Z500) field during
1979–2009 (Fig. 5d). The absence of statistically significant trends
in Z500 over the area can be explained by relatively little change
since the early 1990s (Fig. 5g). The higher geopotential heights
observed over West Antarctica have been described as being part
of an atmospheric Rossby wave train forced by higher SST in the
central tropical Pacific
17
. The signature of this wave train is clearly
apparent in the correlations with Byrd temperature (Fig. 5d), more
in JJA than in any other season. This atmospheric teleconnection is
manifested in the second mode of covariability between tropical SST
(20
◦
S–20
◦
N) and Southern Hemisphere atmospheric circulation
17
.
Notably, it involves an SST forcing distinct from the traditional
eastern equatorial Pacific El Niño/Southern Oscillation (ENSO)
region
17
, yet consistent with the increasing frequency of El Niño
events with SST anomalies in the central Pacific
31
.
In SON and DJF, the trends in Z500 project more poorly onto
the spatial correlations than in JJA (Fig. 5b,e), suggesting a more
complex causality of the warming. In other words, the mechanisms
accounting for the secular temperature trends may differ (in part)
from those responsible for the interannual temperature variability.
In SON, the West Antarctic warming has been primarily attributed
to lower geopotential heights in the South Pacific that have
enhanced northerly warm air advection towards West Antarctica
8
(Fig. 5b). This change in the atmospheric circulation is congruent
with the trends in the two modes of high-latitude atmospheric
variability often associated with ENSO, and known as the Pacific
South American modes
8,32
. These trends, however, have been linked
to positive SST anomalies in the tropical branch of the South
Pacific Convergence Zone
17
(SPCZ) and seem to be more clearly
distinct from ENSO variability (either of the eastern or central
Pacific type) compared with JJA (ref. 17). The close parallel between
SST anomalies in the subtropical SPCZ region and West Antarctic
temperatures over the past 50 years
8
suggests a more southern
location of the relevant SST forcing than indicated in ref. 17. We
confirmed this hypothesis with a covariance analysis between SST
and 200 hPa geopotential heights (Z200) similar to that conducted
in ref. 17, but encompassing SSTs beyond the tropical latitudes (see
Supplementary Discussion and Figs S2 and S3).
The link between West Antarctic warming and (sub)tropical
SST anomalies has not been established convincingly with model
sensitivity experiments in SON (refs 8,17), in contrast to JJA
(ref. 17), suggesting again that other mechanisms may be at play.
Remarkably, the record-high temperatures at Byrd in the mid-
1990s to late 2000s occurred along with a marked increase in the
interannual temperature variability, which is also well reflected in
the Z500 time series over the Bellingshausen Sea sector (Fig. 5h).
NATURE GEOSCIENCE | VOL 6 | FEBRUARY 2013 | www.nature.com/naturegeoscience 141
© 2013 Macmillan Publishers Limited. All rights reserved.
ARTICLES
NATURE GEOSCIENCE DOI: 10.1038/NGEO1671
¬0.6
90 S°
180° 120° W60° W0°
Longitude
60° E 120° E 180°
60° S
30° S
0°
Latitude
30° N
60° N
90° N
Annual DJF MAM JJA SON DJ Annual DJF MAM JJA SON DJ
¬1.0
¬0.5
0.0
0.5
1.0
1.5
2.0
2.5
¬0.3
0.0
°C per decade
0.6
0.3
0.9
1.2
1.5
a
c
°C per decade
b
1958¬2010 trends
1958¬2009 annual temperature change
1980¬2010 trends
0.4
0.8
1.2
1.6
Temperature change (°C)
2.0
2.4
Figure 3 | Linear temperature trends at Byrd. a,b, Linear trends in the annual and seasonal mean temperature at Byrd during 1958–2010 (a) and
1980–2010 (b). DJ denotes the December–January average. The error bars denote the 95% confidence interval. Trends significant above (below) the 95%
level are shown in red (grey). Details on the error calculation are given in the Methods. The trend values and their statistical significance are given in
Supplementary Table S1. c, Annual mean surface temperature change (that is, trend× number of years) during 1958–2009 from our Byrd record (red and
black circle) and from the CRUTEM4 data set
50
(background map).
¬0.9
Annual
This study
Ref. 14
Ref. 6 (v1)
Ref. 6 (v2)
Ref. 7 (RLS)
Ref. 7 (E-W)
DJF MAM JJA SON
¬0.6
¬0.3
0.0
°C per decade
0.6
0.3
0.9
1.2
1.5
1958¬2005 temperature trends at Byrd
Figure 4 | Comparison of the temperature trends at Byrd from several
reconstructions. Linear temperature trends for 1958–2005 from our
reconstructed Byrd record (this study) and from other Antarctic
temperature data sets (refs 6,7,14). For ref. 14, we use the updated version
of its reconstructed Byrd record
20
. Further details about the data sets are
given in Supplementary Table S2, along with the trend values. The error
bars denote the 95% confidence interval and are calculated as explained in
the Methods.
This increased variability can be related to the greater in-phase
behaviour between the Southern Annular Mode (SAM) and ENSO
observed since the early 1990s (refs 33,34). It also explains the
small Z500 trends over the Bellingshausen Sea region (Fig. 5b), by
compensation of large anomalies of opposite signs. Importantly, the
warmest SON at Byrd (in 2002) coincided with an exceptional and
well-documented sudden stratospheric warming over Antarctica,
following an early breakdown of the polar vortex
35
. There is,
however, no evidence of such a phenomenon in other abnormally
warm springs at Byrd (for example, 1999, 2000 and 2005). Thus, one
can assume that this mechanism is probably not a significant driver
of the long-term SON warming, which is further supported by the
fact that the Antarctic polar vortex tended to break up later in the
1990s than in the 1960s (ref. 36).
Hypotheses for the summer warming
Owing to the relative novelty of a West Antarctic warming in
DJF, little has been said about its possible attribution. On the
contrary, studies have generally emphasized the cooling effect of
recent atmospheric circulation changes for West Antarctica in
austral summer
37–39
. Largely dominated by the positive trend in
the SAM index
40
, these changes have been characterized by lower
geopotential heights over the continent
41
(Fig. 5c) and stronger cir-
cumpolar westerlies, reducing the meridional heat exchange. This
negative relationship between the strength of the SAM and West
Antarctic temperatures is apparent in the Byrd temperature–Z500
correlations calculated for 1989–2011 (Supplementary Fig. S4), but
mostly vanishes when the period is extended back to 1979 (Fig. 5f).
142 NATURE GEOSCIENCE | VOL 6 | FEBRUARY 2013 | www.nature.com/naturegeoscience
© 2013 Macmillan Publishers Limited. All rights reserved.
NATURE GEOSCIENCE DOI: 10.1038/NGEO1671
ARTICLES
Z500 trends
SON DJF
JJA
¬0.3
¬0.3
¬0.3
¬0.3
¬0.3
¬
0.4
¬
0.6
0.4
¬0.4
¬0.3
¬0.3
¬0.6
¬0.4
¬
0.6
¬0.4
¬0.3
¬0.3
¬0.3
¬0.3
¬0.5
¬0.5
¬0.3
¬0.3
¬
0.5
¬
0.3
Box 1 Box 2
Box 3
Byrd temp¬Z500 correlation
Z500 trendsByrd temp¬Z500 correlation
JJA
SON DJF
ab
de
Z500 trendsByrd temp¬Z500 correlation
c
f
JJA SON DJF
¬16
¬12
¬8
¬4
0
4
8
Metres per decade
12
16
1980 1990 2000
Year
2010 1980 1990 2000
Year
2010 1980 1990 2000
Year
2010
¬2
Byrd temp. Z500 (box 1) Byrd temp. Z500 (box 2) Byrd temp. Z500 (box 3)
¬3
¬1
0
Std. dev.
Std. dev.
1
2
3
¬2
¬3
¬1
0
1
2
3
Std. dev.
¬2
¬3
¬1
0
1
2
3
gh i
Figure 5 | Relationships between Byrd temperature and the atmospheric circulation during the three warming seasons. a–c, Linear trends in seasonal
mean Z500 from ERA-Interim during 1979–2009. The thick black dashed lines denote the 95% significance level of the trends. d–f, Correlations (r)
between Byrd temperatures and Z500 calculated for 1979–2009. The correlations are performed with detrended time series. Areas with r > 0.5 over West
Antarctica are denoted with a thick red line. Solid thin red contours and dashed blue contours denote positive and negative correlations, respectively.
g–i, Times series of seasonal mean temperature at Byrd (black line) and Z500 averaged over boxes 1, 2 and 3 in d,e and f, respectively (red line). The black
arrows in a–f show the direction of the prevailing warm air advection.
This supports the fact that the observed strengthening of SAM is
inconsistent with a summer warming at Byrd. Moreover, it is clear
from Fig. 5i that the changes in Z500 over the Bellingshausen Sea
region during the 1980s fail to explain the quasi-stepwise increase
in Byrd DJF temperature around 1986–1989.
We found this late 1980s warming to be consistent with a
westward shift and deepening of the persistent centre of low
pressure over the Amundsen/Ross Sea sector (see Supplementary
Discussion and Figs S5 and S6). The resulting anomalous northerly
warm advection towards Byrd was further enhanced by the
adiabatic warming of air masses descending onto the lee side
of Marie Byrd Land’s coastal mountain ranges (föhn effect)
42
.
However, the position of the low—known to be influenced by
ENSO (ref. 43)—does not exhibit any significant trend over the
post-1979 period
44
and, therefore, cannot account by itself for the
long-term summer warming at Byrd. It is also noteworthy that the
warming at Byrd coincided with a sharp decrease in summer sea-ice
concentrations in the Bellingshausen Sea in 1989, which resulted
from the permanent loss of multiyear sea ice
45
. There is, however,
no clear evidence of a linkage between the two phenomena.
The SST region potentially linked to Byrd summer warming
cannot be identified in a straightforward manner, as exemplified
by the two strongly contrasting SST anomaly patterns associated
with peaks in Byrd temperature in DJF 1997–1998 and DJF 2005–
2006 (Fig. 2 and Supplementary Fig. S7). As in the other seasons,
the second mode of covariability between (sub)tropical SST and
Southern Hemisphere atmospheric circulation best captures the
observed SST trends, in particular the warming of the subtropical
SPCZ region
46
(Supplementary Figs S2 and S3). This mode, which
is separate from ENSO variability (as in SON), is thus likely to reflect
the SST forcing responsible for long-term temperature trends at
Byrd. The patterns of anomalies associated with this second mode
show anomalous northerly winds over the Byrd region occurring
in conjunction with higher SSTs over the subtropical SPCZ region
(Supplementary Fig. S2). With this impact on the atmospheric
circulation, the warming of the subtropical SPCZ region may have
at least mitigated the cooling induced by a stronger SAM, and at
most contributed to abnormally high temperatures at Byrd as seen
in DJF 2005–2006 (Supplementary Fig. S7).
Our reconstructed Byrd temperature record reveals one of
the most rapidly warming places on the planet since the 1950s,
and its spatial footprint (Fig. 1) indicates that similar change
has probably occurred over a broad area of West Antarctica.
These results underscore the importance of maintaining a robust
NATURE GEOSCIENCE | VOL 6 | FEBRUARY 2013 | www.nature.com/naturegeoscience 143
© 2013 Macmillan Publishers Limited. All rights reserved.