Introduction
Observations during the second half of the twentieth century suggest
significant but spatially complex variability in atmospheric and ocean
temperature and circulation (Figs. 1 and S1 in the Supplement), as well as
ice-sheet dynamics, across the mid- to high latitudes of the Southern
Hemisphere (Jones et al., 2016). These factors include an intensification of
western boundary currents
(Wu et al., 2012), a strengthening and poleward shift in the summer westerly
winds associated with a positive trend in the southern annular mode (SAM)
(Marshall, 2003; Abram et al., 2014; Thompson et al., 2011), winter–spring
warming over West Antarctica (Steig et al., 2009), latitudinal shifts in the
subantarctic and polar fronts associated with the Antarctic Circumpolar
Current (ACC) (Langlais et al., 2015), spatial and temporal changes in sea
ice extent (Turner et al., 2015; Hobbs et al., 2016), and Antarctic ice-sheet
mass loss (Pritchard et al., 2012). Unfortunately, major uncertainties exist
regarding their trends and interaction(s) due to high interannual variability
(Turner et al., 2016; Fogt et al., 2012) and limited instrumental records
prior to the 1950s (Goosse and Zunz, 2014; Jones et al., 2016). As a result,
analysis has relied on modelling studies to infer multidecadal-to-centennial
variability (Freitas et al., 2015; Wang and Dommenget, 2016) and explore
regional and global teleconnections (Langlais et al., 2015; Goosse and Zunz,
2014), both of which may have changed with anthropogenic forcing. The above
uncertainties are particularly acute in the South Pacific Ocean and adjoining
regions because of the expression of central tropical ocean–atmospheric
interactions associated with the El Niño–Southern Oscillation (ENSO)
(Abram et al., 2014; Schneider et al., 2012; Turney et al., 2016a; Ciasto and
Thompson, 2008; Ding et al., 2012).
Ocean–atmosphere coupling in the Southern Hemisphere.
(a) Significant (p<0.05) austral summer (December–February) sea
surface temperature (SST ∘C decade-1; shading) and 925 hPa winds (vectors) trends since
1979. Temperatures based on SSTs from the HadISST dataset (Rayner et al.,
2003); winds from ERA-Interim (Dee et al., 2011). Key sites discussed in text
are shown: Macquarie Island (MI), Campbell Island (CI), Antipodes Island
(AI), Ferrigno (F), Bryan Coast (BC), Gomez (G) (Thomas et al., 2008, 2015), Falkland Islands (FI), and South Georgia (SG) (Turney et al.,
2016a). Overlaid in green are the three main fronts of the Antarctic
Circumpolar Current (Sallée et al., 2012). (b) Annual
(July–June) and spring–summer (October–March) air temperatures at Macquarie
Island. Dashed lines denote range of the Australasian Antarctic Expedition
temperatures (AAE; CE 1912–1915). Period of satellite
observations (a) shown by grey bar; dashed coloured lines denote
trend in temperatures across the satellite period. (c) Monthly
Macquarie Island air (red line) and sea surface temperatures (blue line)
(with 1σ) demonstrating tight coupling between atmospheric
temperature and SSTs (CE 2000–2014).
Late twentieth century climate over the Southern Ocean is characterised by
high interannual variability (Jones et al., 2016; Turner et al., 2016),
in part driven by changes in the strength and location of mid-latitude
westerly airflow (Thompson et al., 2011). SAM and ENSO play a dominant role
in this as modes of large-scale variability (Fogt et al., 2012; Ciasto and
Thompson, 2008). Of particular significance, the positive post-1960s trend in
the mid- to high-latitude pressure gradient described by SAM reaches its
maximum during the austral summer (Jones et al., 2016), marked by a zonally
symmetric poleward displacement of the jet stream and strengthening of the
prevailing surface westerly air flow centred on 50∘ S (Marshall,
2003; Thompson et al., 2011) (Fig. S2). In contrast, ENSO is associated with
spatially different temperature and wind relationships across mid- to high
latitudes (Ciasto and Thompson, 2008) (Fig. S3), with atmospheric pressure
anomalies experiencing their greatest amplitude during austral spring and
summer in the South Pacific (Fig. S4). The pattern resembles a zonally
asymmetric wave train of atmospheric pressure anomalies extending from New
Zealand to the West Antarctic Coast, and into the Weddell Sea–South Atlantic
(the so-called Pacific–South American or PSA mode) (Mo and Higgins, 1998;
Trenberth et al., 2014, 1998; Karoly, 1989). The PSA has
been shown to introduce zonal asymmetries in the seasonal SAM structure in
the South Pacific (Fogt et al., 2012). Overall, the poleward migration of
storm tracks reduces air-to-sea heat fluxes through increased cloud cover and
evaporative heat loss from the ocean (Thompson et al., 2011; Ciasto and
Thompson, 2008), while increasing oceanic Ekman transport of cool surface
water (Ciasto and Thompson, 2008) and a poleward eddy heat flux (Sallée
et al., 2012). As a result, sector-specific poleward shifts in westerly
airflow have led to contrasting late twentieth century ocean–atmospheric
trends. How the above modes of variability influenced Southern Ocean climate
and ocean dynamics before the period of satellite observations remains highly
uncertain (Jones et al., 2016). An improved network of quantified
climate-sensitive proxy records across the mid- to high-latitudes is crucial
for exploring climate teleconnections through time (Jones et al., 2016; Abram
et al., 2014; Turney et al., 2016a, 2015).
The subantarctic islands of the southwest Pacific lie at hemispherically
important atmospheric and ocean boundaries, offering considerable potential
for understanding long-term climate trends and the potential role of tropical
forcing on high-latitude change. Campbell (52.54∘ S,
169.14∘ E) and Macquarie (54.50∘ S, 158.95∘ E)
islands are located just north of the main front of the ACC and south of the
Subtropical Front (also known as the Subtropical Frontal Zone or Convergence)
(Fig. S6) (Streten, 1988; Sokolov and Rintoul, 2009) in the core latitude of
Southern Hemisphere westerly airflow (Streten, 1988), and are sensitive to
Rossby wave propagation from the tropics to the high latitudes (Adamson et
al., 1988; Ding et al., 2012) (Fig. S4). Campbell and Macquarie islands have
some of the longest, near-complete, continuous instrumental records in the
Southern Ocean (commencing 1941 and 1948 respectively) (Table S1 in the
Supplement) supplemented by daily atmospheric and sea surface temperature
(SST) measurements made at Macquarie Island between common era (CE) 1912 and
1915 as part of Sir Douglas Mawson's landmark Australasian Antarctic
Expedition (AAE) (Kidson, 1946). Mawson's observations span 4 years and
resolve the seasonal cycle, therefore allowing comparison to the continuous
instrumental record from the 1940s to the present day (hereafter “the modern
record”). The time series of observed temperatures on the two islands are
highly correlated in the modern record (detrended July–June correlation
0.801, p<0.0001) and display the same significant spatial correlation
fields to regional and Pacific-wide SSTs (Fig. S5), demonstrating a
comparable climate regime. As a result of anomalies in the overlying wind,
the surrounding waters are strongly influenced by variations in northward
Ekman transport of cold fresh subantarctic surface water and anomalous fluxes
of sensible and latent heat at the atmosphere–ocean interface. This has
produced a cooling trend since 1979 (Figs. 1 and S1) (Thompson et al., 2011;
Ciasto and Thompson, 2008), making the islands ideally placed to detect
wind-driven changes in the north–south SST gradient over time.
Here we extend the instrumental record by exploiting the climate sensitivity
of the southernmost-growing trees in the subantarctic southwest Pacific to
produce the first annually resolved quantified temperature reconstruction for
the region back to CE 1870. The maritime climates of Campbell and Macquarie
islands provide a reconstruction of air and sea temperatures that
demonstrates increasing variance since modern records commenced in the
∼ 1940s. We investigate this time series with climate reanalysis and a
three-dimensional Earth system model of intermediate complexity and identify
the tropical Pacific sea surface temperatures as the principal driver of the
observed variance, propagated by atmospheric Rossby waves during the austral
spring and summer. Climate-sensitive records across the circum-Pacific
demonstrate comparable trends, suggesting that tropical climate changes have
been increasingly projected onto the mid- to high latitudes. Subantarctic
islands across the wider Southern Ocean provide crucially situated landmasses
from which proxy data can be generated to test hypotheses about past and
future global climate teleconnections.
Methods
Subantarctic island climate datasets (Macquarie and Campbell
islands)
The AAE 1912–1915 atmospheric
observations were taken from the isthmus at the northern end of Macquarie
Island (Newman, 1929), in the same immediate area as the current Australian
Antarctic Division Meteorological Station, established in late 1948
(54.50∘ S, 158.95∘ E). The daily and monthly meteorological
data from Macquarie Island were obtained from the reduced AAE dataset
(Newman, 1929) and since 1948, the Bureau of Meteorology
(http://www.bom.gov.au/climate/data-services/). Twice-daily SST
measurements were also taken from Buckles Bay during the AAE (Newman, 1929),
with subsequent observations made again during the 1950s and 1960s (Loewe,
1968, 1957); unfortunately no direct measurements exist between 1916–1950.
The next available continuous SST observations that can be compared to the
1912–1915 record are remote MODIS satellite (Terra) measurements providing 4 km-resolved
11 µm daytime observations since 2001 (data accessed
via https://modis.gsfc.nasa.gov/); other satellite products do not
resolve at a spatial scale that allows a direct comparison to the localised
measurements made at Macquarie Island. Although the satellite data are from a
larger area than the AAE observations, the expedition vessel the S.Y.
Aurora made SST measurements across Buckles Bay and demonstrated
similar absolute values as those observed inshore, providing confidence that
the comparisons are robust (Kidson, 1946). A meteorological station has
operated in Perseverance Harbour, Campbell Island (52.54∘ S,
169.14∘ E), since 1941. The dataset used here was obtained from the
New Zealand National Climate Database (http://cliflo.niwa.co.nz/). Near-complete instrumental records have been maintained on Campbell and Macquarie
islands since observations began with no complete months missing from any of
the datasets (Table S1).
Developing a temperature-sensitive tree-ring record from the
subantarctic Pacific. (a) Dracophyllum raw tree-ring
chronology (green line) with different standardisation outputs (various
coloured lines), expressed population signal (EPS; thick red line) and sample
size of trees (blue area). Bootstrap correlation function of the
Dracophyllum tree-ring chronology to instrumental records of monthly
temperatures from Campbell Island (b) and Macquarie Island
(c) with error statistics for early (CE 1949–1980) and late
(1981–2012) calibration periods. Darker bars indicate months with
statistically significant correlations (p<0.05).
Meteorological observations
To extend the satellite record for the southwest Pacific, we focused on the
subantarctic Macquarie and Campbell islands. For comparison to the AAE
1912–1915 record, modern-day Macquarie Island temperature measurements were
compared in 4-year bins (Tables S2 and S3). The interannual variability in
the most complete dataset (that from Macquarie Island) is relatively large.
Student's t tests (two-tailed) of the 4-year average monthly data
relative to 1912–1915 indicate that the most consistently warmer conditions
are during February–April (Tables S2 and S3). This analysis illustrates a
trend towards seasonally restricted warming only during the late austral
summer and autumn. Intriguingly, no pervasive warming is observed across the
austral spring and most of the summer when ENSO and SAM are today known to
play a dominant role on regional climate variability (Ciasto and Thompson,
2008).
Tree-ring reconstruction (dendrochronology)
To develop an annually resolved temperature reconstruction for the southwest
Pacific that will extend the modern instrumental record we sampled 30
Dracophyllum spp. trees from Campbell Island during 2013 as part of
the Australasian Antarctic Expedition 2013–2014, and during further
fieldwork in late 2014 (Fig. 2). Here two Dracophyllum species
(D. longifolium, D. scoparium and hybrids) form the
southernmost growing evergreen shrubs and small trees in the southwest
Pacific (with no Dracophyllum on Macquarie Island) (Wilmshurst et
al., 2004; Turney et al., 2016b). Dracophyllum spp. are known to be
responsive to warmer temperatures and capable of reaching ages of
> 200 years (Harsch et al., 2014), providing an opportunity to derive a
continuous proxy record of temperature in this key region spanning more than
a century. Because of the coherent climate trends on both islands, the
relationship of tree-ring growth to Campbell and Macquarie Island
temperature records was explored using bootstrapped correlation function
analysis in the bootRes R software package (Zang and Biondi, 2012) to
identify the monthly temperature responses, followed by a split period for
calibration/verification analysis to test the regression model robustness
using the reduction of error (RE) and the coefficient of efficiency (CE)
(Fig. 2 and Table S4). Based on those results, we selected an austral
“growing season” window for linear regression modelling to produce
spring–summer (October–March) temperature reconstructions for Campbell and
Macquarie islands.
140-year temperature variability in the subantarctic Pacific.
Campbell Island (a) and Macquarie Island (b) observed (red
lines) and Dracophyllum reconstructed (black) growing season
temperatures (October–March) with 90 % quantile limits (grey envelope)
compared against running 30-year mean standard deviation of the reconstructed
temperature series (c). (d) Box-and-whisker plots of the
ring width indices with summary statistics indicating a significant
difference in variance between the periods CE 1870–1941 and 1941–2012.
Orange column defines significant post-1940s temperature variability in the
record.
After crossdating and measuring, the 30 tree series were standardised to
remove biological trends using the RCSigFree program
(http://www.ldeo.columbia.edu/tree-ring-laboratory/resources/software).
Within the program, various options are available for the conversion of the
annual ring width measurements into indices and we adopted the use of a more
flexible regression model, the Friedman Super Smoother (Friedman, 1984), to
remove the growth trends. The ring width measurements were first power
transformed and then subtracted from the regression model to produce indices
and avoid possible outlier bias (Cook and Peters, 1997). Following this, the
signal-free method was applied to minimise trend distortion and end-effect
biases in the final chronology (Fig. 2) (Melvin and Briffa, 2008). Comparison
between the detrended series and average raw measurements (Fig. 2)
demonstrate the standardisation process (or any of the other models) did not
make the series heteroscedastic. The relationship of the tree-ring chronology
to the Campbell and Macquarie Island temperature records and Southern
Annular Mode (SAM) reconstruction (Visbeck, 2009) was explored using
bootstrapped correlation function analysis in the bootRes R software package
(Figs. 2 and S17) (Zang and Biondi, 2012). bootRes uses 1000 bootstrapped
samples to compute Pearson's correlation coefficients between the tree-ring
parameter and each of the climatic predictors and then to test their
significance at the 0.05 level. Bootstrap samples are drawn at random with
replacement from the selected time interval. Median correlation coefficients
are deemed significant if they exceed, in absolute value, half the difference
between the 97.5th quantile and the 2.5th quantile of the 1000 estimates
(Biondi and Waikul, 2003). In the plots, the darker bars indicate a
coefficient significant at p<0.05 and the lines represent the
95 % confidence interval.
Based on these results, the Campbell Island “growing season” of monthly
temperatures from October to March (six months, spanning from spring to
autumn) was selected for reconstruction using the Dracophyllum
chronology for the period 1870–2013 (Expressed Population Signal or
EPS > 0.85). Similarly for Macquarie Island, the same growing season was
selected (October–March; six months). For SAM, we find the most significant
relationship was for July–October. The program PCReg
(http://www.ldeo.columbia.edu/tree-ring-laboratory/resources/software)
was used to carry out a linear regression model of the tree-ring chronology
to the selected growing-season windows for both Campbell and Macquarie
islands. A split period for calibration/verification analysis was used (Cook
and Kairiukstis, 1990) to test the regression model robustness. Our model for
Campbell Island passed both the CE and RE tests (i.e. positive) indicating
that the model was skillful in reconstructing observed variations, however
the verification results for Macquarie Island were weaker and just failed for
the more rigorous CE test (Table S4). We then used the full period of
instrumental data (1949–2012 for Campbell Island and Macquarie Island) to
develop final models and reconstruct “growing-season” temperatures back to
1870 for both islands (Fig. 3). The prediction intervals (90 % quantile
limits) associated with the reconstructed temperatures were produced using a
fixed t statistic for scaling the uncertainties (Olive, 2007). Importantly,
the chronology is derived from a mixture of tree ages (i.e. the oldest
started in CE 1747 and the youngest in 1958) and is not made up of a single
cohort of similar aged trees that have matured across the same period.
Tropical variability in the subantarctic temperature record.
Changing amplitude of reconstructed summer temperatures for Macquarie Island.
Multi-taper method (MTM) (a) and extracted climate periodicities
exceeding 99 % significance (b) observed in the
Dracophyllum-derived growing season temperature reconstruction for
Macquarie Island since CE 1870.
Spectral analysis
To investigate climate periodicities we undertook multi-taper method (MTM)
analysis on the Dracophyllum temperature reconstruction (Fig. 4),
tree chronology (Fig. S12) and annual southwest Pacific SSTs (Fig. S13) (the
latter derived from Hadley Centre Ice and Sea Surface Temperature; HadISST)
(Rayner et al., 2003) using a narrowband signal and red noise significance
(with a resolution of 2 and 3 tapers) (Thomson, 1982) with the software
kSpectra version 3.4.3 (3.4.5).
Characterising water mass sources and ocean fronts
In order to characterise the decadally averaged source(s) of water masses
near Macquarie and Campbell islands, we performed an experiment with virtual
particles in an eddy-resolving ocean model (the Japanese Ocean model For the
Earth Simulator or OFES) (Masumoto et al., 2004), which has a 1/10∘
horizontal resolution and near-global coverage between 75∘ S and
75∘ N, and has a demonstrated ability for modelling changes in the
Southern Ocean between 2000 and 2010 (van Sebille et al., 2012) (Fig. S6).
While OFES precludes us from modelling the warming across the 1970s, it does
allow us to hindcast the origin of the waters down to a depth of 400 m using
Lagrangian analysis in the most recent decade. Assuming a steady-state ocean
circulation, this analysis allows us to refine our understanding of the
sources and by association boundaries of water masses surrounding Macquarie
and Campbell islands. The model was forced using the National Centers for
Environmental Prediction (NCEP) wind and flux fields and output is available
as 3-day averages (Qin et al., 2014).
Particles were released every three days between 1 January 2005 and
31 December 2010 on a latitude–depth section at 170∘ E, every
0.1∘ in latitude between 60 and 45∘ S and every 50 m in
depth between 25 and 300 m, for a total of 318 288 particles. The particles
were then advected backwards in time within the three-dimensional OFES
velocity fields using the fourth-order Runge–Kutta method as implemented in
the Connectivity Modeling System (CMS) version 1.1b (Paris et al., 2013). The
particles were advected for 5 years, or until they reached 30∘ S
or 0∘ E. Once all the particles were integrated, they were
categorised into those that start in the Agulhas Current (at 30∘ S
and between 28 and 40∘ E) and those that start in the East
Australian Current (at 30∘ S and between 150 and 160∘ E).
Using the western Indian Ocean boundary Agulhas Current as a tracer for the
Subtropical Front (and the southern limit of the Subtropical Gyre) (Wang et
al., 2014) we identify a pathway of particles flowing from the Cape of Good
Hope to the southwest Pacific subantarctic islands (Fig. S6). The particles
that connect to the region around Macquarie and Campbell islands follow a
very narrow and almost linear path southeastward across the Indian Sector of the
Southern Ocean. The fastest particles reach Macquarie Island less than 2 years after release in the Agulhas Current, with the majority arriving
between 3 and 4 years after release. In contrast, little leakage from
the East Australian Current is observed, with approximately 6 times more
Agulhas particles delivered to the southwest Pacific subantarctic than the
East Australian Current (EAC) (Wu et al., 2012) (Fig. S6).
Modelling transient change
General circulation models (GCMs) involved in the Fifth Coupled Model
Intercomparison Project (CMIP5) (Taylor et al., 2011) struggle to simulate
the observed internal variability and/or seasonal cycle over the Southern
Ocean (Wang et al., 2015; Zunz et al., 2013), supported by the poor
correlations observed between our reconstructed and CMIP5 October–March
temperatures (Table S5). Here we take an alternative approach using
LOVECLIM1.3, a three-dimensional Earth system model of intermediate
complexity (Goosse et al., 2010) that includes representations of the ocean
and sea ice (CLIO3) (Goosse and Fichefet, 1999), atmosphere (ECBilt2)
(Opsteegh et al., 1998), and vegetation (VECODE) (Brovkin et al., 2002). The
three-level quasi-geostrophic atmospheric model has a horizontal resolution
approximating 5.6∘ × 5.6∘ (T21) whilst the ocean
general circulation model is coupled to a sea-ice model with 20 unevenly
spaced vertical levels and a horizontal resolution of
3∘ × 3∘. The vegetation component simulates the
evolution of grasses, trees and desert, with the same horizontal resolution
as ECBilt2. The experiments analysed here cover the period CE 1850–2009,
driven by the same natural (solar and volcanic) and anthropogenic (greenhouse
gas, sulfate aerosols, land use) forcings (Goosse et al., 2006) as the ones
adopted in the historical simulations performed in the framework of CMIP5
(Taylor et al., 2011). The initial conditions are derived from a numerical
experiment covering the years CE 1–1850 using the same forcing, in order to
take into account the long memory of the Southern Ocean (Goosse and Renssen,
2005). For the CE 1850–2009 simulations, the model was forced to follow the
observations of surface temperature from the HadCRUT3 dataset (Brohan et al.,
2006) using a data assimilation technique based on particle filtering (Goosse
et al., 2006; Dubinkina and Goosse, 2013). A simulation without additional
freshwater flux (no freshwater flux) with data assimilation, from CE 1850 to
2009, was analysed here (Zunz and Goosse, 2015), allowing direct comparison
between climate parameters and SST trends across the Southern Ocean. SSTs for
the Macquarie–Campbell Island sector and anomalies in zonal wind stress are shown in Figs 8 and 9 respectively.
Results and discussion
Modern climate changes
Comparing atmospheric temperatures during the 1912–1915 AAE observational
period and the modern record from Macquarie Island demonstrates high
interannual variability (Fig. 1b). Whilst the temperature trend across the
period of satellite observations appears to show a cooling trend in the
southwest Pacific, significant warming is observed across the annual and
spring–summer months from the 1960s and peaks during the 1980s (Figs. 1 and
S7, Table S2). No parallel changes are observed in wind direction (Fig. S8)
while the sunshine time series appears to trend in the opposite direction to
that expected (Fig. S9). The number of ocean observations are more limited,
but comparable warming (0.5 ∘C) was observed across the 1950s–1960s
with MODIS satellite measurements (MODerate Imaging Spectroradiometer;
2000–2014) demonstrating slightly cooler waters during the present day
(though still 0.3 ∘C warmer than the AAE period) (Table S3). A
similar long-term trend is also observed with air and sea temperatures at
Campbell Island (Morrison et al., 2015). Importantly, because of their small
size and highly maritime climate, atmospheric temperatures on the islands
parallel the seasonal SST cycle (Fig. 1c), indicating a tight thermal
coupling between air and sea surface temperatures (Thompson et al., 2011;
Kidson, 1946; McGlone et al., 2010), providing a sensitive terrestrial
measure of Southern Ocean conditions.
Changing climate variability
The Dracophyllum reconstructions extend the surface air temperature
record for the southwest Pacific sector of the Southern Ocean back to
CE 1870 (Fig. 3). We find highly variable growing-season (spring–summer)
temperatures that parallel meteorological observations on the subantarctic
islands for the period of overlap (including the original AAE) (Fig. 3), with
a trend towards increasing temperatures from the 1960s that reached a maximum
during the late 1980s (∼ 1 ∘C warmer on Macquarie Island
compared to period 1912–1915). Peak temperatures of the 1980s, however, were
not sustained in the southwest Pacific through to present day (Fig. 1).
Instead, a notable feature of our 140-year reconstruction is the long-term
change in variability captured by a 30-year running standard deviation,
regardless of the standardisation method used (Figs. 3c and S11). We observe
a sustained increase from the ∼ 1940s compared to intermediate levels
of variance during the late nineteenth century and a minimum during the first
half of the twentieth century. The high number of replicated trees across the
reported series means we can discount changing sample depth as the cause of
increasing variance. Removing extreme values centred on 1956, 1979, and 1986
does not substantially change the shift to higher variance in the second half
of the twentieth century (Fig. S10), demonstrating that the long-term trend
is robust. To test for the significance of this change, we compared the
variance across the tree-ring record (CE 1870–1941 vs. 1941–2012) and
found the second half of the twentieth century is significantly larger for
all standardisation approaches (Friedman F and Bartlett's K squared tests p=0.0055; Table S6), suggesting a shift in climate to one characterised by
pervasive higher variability.
To further investigate the change in temperatures across the record we
undertook multi-taper method (MTM) spectral analysis on the reconstructed air
temperature and associated tree-ring index (Figs. 4 and S12). We find the
strongest periodicities in growing season temperatures over two narrow
windows, 3.1 and 2.4 years (all above 95 % confidence), identical to those
recognised in regional SSTs extracted from HadISST (Rayner et al., 2003)
(Fig. S13). Hovmöller plots of satellite-observed SSTs between 45 and
55∘ S confirm a pattern of alternating warm and cool temperatures in
the southwest Pacific subantarctic islands with these periodicities
(Fig. S14). Our new extended temperature series therefore indicates the late
nineteenth and early twentieth century climate was characterised by low
interannual variability with increasing amplitude in the 3.1 and 2.4-year
bands from the ∼ 1940s and late 1960s respectively (Fig. 4b). Recent
work by Chelton and Risien (2016) suggest that there is an increase in
standard deviation in HadISST from 1949. Our tree-ring temperature
reconstruction, however, shows a real variance increase that is independent
of this artifact in the observational data. We therefore conclude the
increased amplitude of the 3.1 and 2.4-year bands is a robust climate feature
in the southwest Pacific since the 1940s.
Marine population changes
Recent work has illustrated how multi-stressors (including climate
variability) can impact on Southern Ocean biota (Boyd et al., 2015) and have
potentially dramatic biological responses across different trophic levels
(Trathan et al., 2007; Constable et al., 2014), including reduced breeding
success (Lea et al., 2006). Intriguingly, the observed increase in variance
reported here appears to coincide with a regional order of magnitude decline
in the populations of many marine species across the southwest Pacific
(spanning Macquarie Island to the Antipodes Islands), including penguins and
elephant seals (Weimerskirch et al., 2003; Morrison et al., 2015;
Childerhouse et al., 2015; Moore et al., 2001; Baker et al., 2010). Top
marine predators can provide an integrated view of an ecological system,
offering a measure of the impact of climate changes on the availability of
food supplies (abundance and distribution), and on feeding and breeding
habitats (Jenouvrier et al., 2003). Whilst not a focus of the current study,
the following provides a brief summary of penguin and elephant seal
population trends as a basis for comparison to the climate and ocean trends
and variability reported here.
In the New Zealand subantarctic there have been pronounced declines in the
numbers of eastern rockhopper penguins (Eudyptes filholi) at
Campbell Island, and both rockhopper and erect-crested penguins (E.
sclateri) on the Antipodes Islands (49.68∘ S,
178.75∘ E) (Table S7). On Campbell Island, the 1940s breeding
population of rockhopper penguins was estimated at 1.6 million birds,
declining through the 1950s followed by a brief resurgence in numbers, before
a further decline that began no later than the mid-1970s (Cunningham and
Moors, 1994). By 2012, rockhopper numbers on Campbell Island had suffered a
95.5 % decline (of which 94 % had occurred by the mid-1980s)
(Morrison et al., 2015). Allowing for a lag of several years for chicks to
reach breeding age, the changes in rockhopper penguin numbers correlate with
changes in sea water temperatures recorded in Perseverance Harbour which
increased to a peak between 1945 and 1950, declined between 1950 and 1965,
then increased sharply by 1970 (Morrison et al., 2015; Cunningham and Moors,
1994). For the Antipodes, data on the decline in both eastern rockhopper and
erect-crested penguin populations cover a shorter period, but are more
robust. Whole-island group surveys have been conducted on three occasions
and, although there were some differences in counting methodology and time of
year in which counts were made, the decline in both species has been
substantial; in 2011 there were only about 5 % as many rockhopper
penguins and fewer than half as many erect-crested penguins as there were in
1978 (Table S7) (Hiscock and Chilvers, 2014). Whilst no climate data are
available from the Antipodes Islands, this subantarctic archipelago falls
within the same climate zone as Macquarie and Campbell islands (Fig. 1) and
is therefore assumed to have experienced the same long-term trend in air and
sea surface temperatures.
Land-based threats do not account for the declines observed. Nesting habitat
availability is unchanged and introduced mammals are not generally considered
to pose a threat. On Campbell Island, Norway rats (Rattus
norvegicus) were present until
eradicated in 2001, while feral cats (Felis catus) died out naturally between 1979 and 1999. However, rats are thought to only prey on eggs once they
are broken through other causes and there was no evidence to suggest that the
few cats present preyed on rockhopper penguins, their eggs or chicks
(Cunningham and Moors, 1994). Avian cholera was recorded in Campbell Island
rockhopper penguins in 1885/86 and 1986/87, but the numbers killed do not
account for the magnitude of the declines recorded (Cunningham and Moors,
1994). Feral sheep (Ovis aries) were present (since eradicated) but
penguin numbers declined in both accessible and inaccessible colonies
(Cunningham and Moors, 1994). On the Antipodes Islands, house mice (Mus musculus) are the only introduced mammal and they are too small to pose a
threat to penguins.
Similar to penguin populations, the number of elephant seals
(Mirounga leonina) have also declined on both Campbell and Macquarie
islands since the 1940s (McMahon et al., 2005) with the decrease being most marked on Campbell
Island which is further from the polar front (Antarctic Convergence),
considered to be the optimum foraging habitat for the species (Taylor and
Taylor, 1989). Pup production on Campbell Island declined from 191
individuals in 1947, 11 in 1984, to just five in 1986 (Taylor and Taylor,
1989; McMahon et al., 2005), with the population falling from 455 before the 1970s to less than 10 individuals in the 2000s (McMahon et al., 2005). On Macquarie Island, the population decreased from around 183 000 in 1949 to some 76 000 in 2001 (Hindell and Burton, 1987), and has since apparently stabilised (van den Hoff et al., 2007). The most likely explanation for those
declines are decreases in marine food availability due to changes in the
marine environment.
Because of the scarcity of island breeding sites and their limited foraging
range while breeding, subantarctic penguins are particularly susceptible to
climate change and associated changes in marine parameters. Penguins,
elephant seals, and other top predators may respond to changes in food
availability when marine parameters change by retracting or expanding their
distributions, with changes in population size or breeding phenology
(Weimerskirch et al., 2003; McMahon et al., 2005). Alternatively, climate change can affect
populations due to changes in conditions ashore. For example, at Punta Tombo
in Argentina since 1960 storms have become more frequent and more intense
causing the deaths of Humboldt penguin (Spheniscus humboldti) chicks
(Boersma and Rebstock, 2014). At Punta Tombo, chick deaths due to storms were
additive to deaths due to other factors. It is important to note, however,
that there is usually a lag between climate change and any subsequent change
in penguin (or other predator) population; the lag time depending on whether
climate affects adult or chick survival, recruitment or some other
demographic parameter (Weimerskirch et al., 2003). Future work is now needed
to investigate this relationship further and identify which changes in marine
parameters may be the cause.
Correlations and significance of relationship between subantarctic
island air temperatures and measures of regional (50–60∘ S,
150–170∘ E) and equatorial Pacific sea surface temperature (SST)
and atmospheric circulation. Regional and Nino temperature anomalies as
calculated from HadISST (Rayner et al., 2003); the Southern Oscillation Index
(SOI) as reported by Ropelewski and Jones (1987). Deseasonalised and
detrended correlations derived for the period CE 1979 to 2014. Significance
indicated as follows: a p<0.05, b p<0.01, and
c p<0.001 (latter given in bold).
SW Pacific
Nino 4
Nino 3.4
Nino 3
Nino 1 + 2
SOI
Macquarie Is.
July–June
0.813c
-0.392a
-0.512b
-0.563c
-0.558c
0.470b
October–March
0.835c
-0.396a
-0.523b
-0.596c
-0.614c
0.475b
Campbell Is.
July–June
0.754c
-0.412a
-0.514b
-0.546c
-0.531c
0.458c
October–March
0.782c
-0.409a
-0.534b
-0.592c
-0.582c
0.473b
Climate controls on temperature over Macquarie Island. Spatial
correlations between detrended and deseasonalised Macquarie Island mean
monthly atmospheric temperatures (October–March) and 850 hPa
height (a), zonal wind stress using ERA-Interim31 (b)
and sea surface temperature (HadISST; c) (Rayner et al., 2003) for
the period 1979–2013 (pfield<0.05). Note: Campbell Island (CI)
and the Antipodes Islands (AI) fall within the region of greatest correlation
to SSTs in the southwest Pacific. The southwest Pacific (SW Pacific;
50–60∘ S, 150–170∘ E) and Nino 3 regions also shown. For
comparison, mean seasonal (October–March) daily wind run (kilometres) for
the meteorological station at Macquarie Island (source: Bureau of
Meteorology) with comparison to average from the Australasian Antarctic
Expedition (1912–1915) with 1σ uncertainty (d). Note, the
period of decreased wind speed across the 1980s coincides with maximum air
temperatures over Macquarie Island (see Fig. 3).
Investigating ocean–atmosphere teleconnections
Whilst the southwest Pacific subantarctic islands lie along the northern edge
of the ACC and south of the Subtropical Front (Fig. S6), the absence of
propagating SST signals across the Southern Ocean suggests that movement of
ocean boundaries and/or changing input of marine western boundary currents
(Figs. S6 and S14) are not primary drivers of the observed increased
variability. An alternative scenario for the increasing amplitude in the 3.1
and 2.4-year bands is a change in atmospheric circulation. To identify a
possible atmospheric mechanism, we compared air temperatures over Macquarie
Island with estimates from ERA-Interim reanalysis (Dee et al., 2011) and
observe a significant positive correlation to spring–summer atmospheric
pressure anomalies (deseasonalised and detrended at 850 hPa) since 1979
(Fig. 5a) and inverse relationships with temperature and zonal and meridional
wind stress (Figs. 5b and S15). Cooler temperatures over Macquarie Island are
therefore associated with a centre of relatively low pressure (at 850 hPa)
south of New Zealand and enhanced westerly and southerly airflow across a
longitudinal band spanning 120 to 150∘ E (significance
pfield<0.05). A similar positive correlation to spring–summer SSTs
is observed with both Macquarie Island (Fig. 5c) and Campbell Island
(Fig. S5), with highly significant relationships to a sector in the southwest
Pacific (50–60∘ S, 150–170∘ E; Table 1), supporting our
earlier observation of the thermal coupling between atmospheric and ocean
temperatures but extending across the broader region. Although we find no
evidence for a sustained shift in airflow direction that parallels the
observed trend in subantarctic temperatures (Fig. S8), we do observe a marked
increase in wind strength across the late twentieth century, with a long-term
intensification (with high variability) of winds that closely parallels air
temperatures over Macquarie Island (Fig. 5d); the original AAE data are
plotted for completeness but given uncertainties over the reliability of
historic observations (Jakob, 2010) a direct comparison is not possible. This
trend towards stronger winds is accompanied by an increase in sunshine hours
over Macquarie Island (Fig. S9), consistent with reduced cloud cover, but any
associated increase in sensible heat flux appears to be offset by increased airflow over cooler surface waters in the southwest
Pacific (Thompson et al., 2011). Our results, therefore, are in line with the
observed (post-1979) spring–summer trend towards windier conditions in the
southwest Pacific (Fig. S1).
Whilst some studies have suggested a dynamical atmospheric circulation
response to ozone layer depletion over the Southern Hemisphere mid-latitudes
since the 1990s (Thompson et al., 2011), the reconstructed 3.1 and 2.4 year
periodicities suggest a tropical teleconnection with the southwest Pacific
(Kestin et al., 1998; Adamson et al., 1988). Using the HadISST (Rayner et
al., 2003) and ERA-Interim (Dee et al., 2011) datasets, a significant inverse
correlation is observed between subantarctic and central–eastern low-latitude
Pacific temperatures and zonal wind stress, with a relatively warm (cool)
eastern equator associated with weaker (stronger) mid-latitude westerly winds
and cooler (warmer) SSTs in the southwest Pacific (Figs. 5a–c and S5).
Comparison to different measures of tropical Pacific SSTs and atmospheric
circulation indicate the most significant relationship with subantarctic
spring–summer temperatures is the Nino 3 region (correlation -0.592, p<0.001) (Table 1).
Rossby wave propagation from the tropical Pacific during the austral
spring–summer. Low-to-high-latitude atmospheric teleconnections during the
austral spring and summer (October–March). Schematic showing extratropical
Pacific–South America (PSA) Rossby wave train (red arrows) associated with
low- and high-pressure systems generated by anomalous equatorial upper-level
divergence flow (Trenberth et al., 1998); enhanced southerly airflow across
the West Antarctic coastline extends into the South Atlantic during
anomalously high temperatures in the Nino 3 region (a). Spatial
correlations between detrended and deseasonalised Nino 3 sea surface
temperature (Rayner et al., 2003) (October–March) and 850 hPa
height (b) and zonal wind stress (c) using ERA-Interim (Dee
et al., 2011) for the period 1979–2015. Location of key sites are shown.
Significance pfield<0.05.
To elucidate the mechanism by which changes in the tropical Pacific may be
projected onto the high latitudes, we explored the relationship between
Nino 3 temperatures and Southern Hemisphere atmospheric circulation using
data from ERA-Interim (Dee et al., 2011) (Fig. 6). We observe what appears to
be a Rossby wave train similar to the PSA climate mode of variability during
the austral spring-summer (Ding et al., 2012; Mo and Higgins, 1998; Trenberth
et al., 1998). We find that post-1979, warmer temperatures in the Nino 3
region leads to deep convection and upper-level divergence flow (at 300 hPa)
(Fogt et al., 2012; Ding et al., 2012; Trenberth et al., 1998) (Fig. S16),
apparently forcing an atmospheric Rossby wave train southeast into the
extratropics manifested as cyclonic anomalies south of New Zealand –
consistent with the relationship observed with Macquarie Island temperatures
(Fig. 5) – that extend across the Pacific as anticyclonic anomalies in the
Amundsen–Bellingshausen seas and cyclonic anomalies off the east coast of
South America (Ciasto and Thompson, 2008; Mo and Higgins, 1998). Lead–lag
analysis demonstrates the atmospheric signal propagates over southern New
Zealand during the late austral winter and reaches the
Amundsen–Bellingshausen seas by the summer (Fig. S17). Our results support
previous studies that find the PSA signal precedes peak temperatures by
approximately one season and abruptly weakens during the austral summer
(Schneider et al., 2012) (Figs. S4 and S17).
Changing Southern Annular Mode (SAM) relationships through the
twentieth century. Running 30-year correlations (a) and bootstrap
correlations (b, c) between hemispheric-wide and sector-specific SAM
reconstructions (July–October) (Visbeck, 2009) and the Dracophyllum
series. Bootstrap correlation periods obtained by halving the SAM dataset
spanning 1890 to 2000. The dark bar indicates that only the Australasian SAM has a
statistically significant correlation to the temperature-sensitive tree-ring
series during the post-1940s period for the austral winter and early spring
(p<0.05).
Modelled changes in Southern Hemisphere westerly airflow over the
last century. Differences in zonal October–March wind speed (m s-1) at
800 hPa across the Southern Ocean derived from LOVECLIM1.3 (Zunz and Goosse,
2015) (a–c) and ERA-Interim (Dee et al., 2011) (d).
Locations of key sites discussed in the text are also shown.
Equatorial and South Pacific temperature and marine population
trends since CE 1860. Nino 3 temperature (July–June) with running 30-year
mean standard deviation of the HadISST temperature series (Rayner et al.,
2003) (a) compared to Campbell Island and Macquarie Island running
30-year mean standard deviation of the reconstructed temperature
series (b). Orange column denotes twentieth century temperature
variability that exceeds any other period in the record. Onset (solid line)
and continuing (dashed) period of declining rockhopper penguin and elephant
seal populations in the southwest Pacific (Morrison et al., 2015;
Weimerskirch et al., 2003) (c) shown for comparison. In addition,
mean annual temperature (∘C, July–June) sea surface temperatures
(HadISST and LOVECLIM model output) for the Campbell–Macquarie islands
region (d) and wider Southern Ocean (e). Note the
coincident increase in West Antarctic Coast (Gomez) (annual and 30-year mean
standard deviation) (Thomas et al., 2015, 2008) (g) and South
Atlantic (Falkland Islands and South Georgia) precipitation (Turney et al.,
2016a) (f).
With the above tropospheric pressure changes (Fig. 6) we suggest that warmer
Nino 3 temperatures are associated with stronger westerly airflow over the
southwest Pacific subantarctic islands and west Antarctic coast, accompanied
by enhanced southerly airflow across the Antarctic Peninsula that extends
into the South Atlantic. Hovmöller plots show an alternating pattern of
warm–cold surface temperatures between the southwest Pacific and
Amundsen–Bellingshausen seas using both the HadISST (Rayner et al., 2003) and
Reynolds v2 SST (Smith and Reynolds, 2005) datasets (Fig. S14), consistent
with atmospheric Rossby wave propagation and regional ocean surface
responses. Running 30-year correlations between the Dracophyllum
series and measures of westerly airflow, however, suggests no relationship
with a hemispheric-wide reconstruction of SAM that extends back to CE 1884
(Fig. 7a) (Marshall, 2003; Visbeck, 2009). Regional monthly changes in the
structure of SAM are now recognised and allow sector-specific analysis (Fogt
et al., 2012; Ding et al., 2012; Visbeck, 2009). Here we identify a
significant inverse correlation to the Australasian region for the austral
winter and spring during the post-1940s period (p<0.05; Fig. 7c), while
the Southern Hemisphere-wide and regional South American and African SAM
reconstructions do not appear to be significant for any period across the
twentieth century (Fig. 7b and c). Previous work has demonstrated that the
PSA is an important contributor to the zonal asymmetry in SAM (Fogt et al.,
2012; Ding et al., 2012), suggesting that the tropics are indeed imposing a signal
on mid-latitude westerly airflow in the southwest Pacific. However, in
contrast to earlier studies which have postulated that anthropogenic forcing may
have changed the structure of SAM to be more zonal (Fogt et al., 2012), our
results imply that the tropics have introduced an asymmetry to the Australasian
sector of SAM in the modern record, or this has at least become more common
during the second half of the twentieth century.
To investigate whether the changes in the southwest Pacific subantarctic
region are representative of a larger part of the Southern Hemisphere we
analysed simulations with the three-dimensional Earth system model of
intermediate complexity LOVECLIM1.3 for CE 1850–2009, driven by natural
(solar and volcanic) and anthropogenic (greenhouse gases, sulfate aerosols,
land use) forcings (Zunz and Goosse, 2015) (Fig. 8). For the 1850–2009
simulation, the model was forced to follow the observations of surface
temperature. We examined the changes in zonal wind stress between selected
decades across the twentieth century, including 1910–1919 (capturing the
original AAE period) (Fig. 8). Over the past century, we find increasingly
stronger westerly winds across the Southern Ocean with a marked
intensification in the southwest Pacific and Antarctic Peninsula during the
most recent decades with more easterly airflow over the Ross Sea (Fig. 8c),
trends also observed in estimates derived from the ERA-Interim dataset
(Fig. 8d) (Dee et al., 2011), and consistent with the observational record
from Macquarie Island (Fig. 5d).
Pacific-wide changes
Although there appears to have been a long-term strengthening of westerly
winds across key sectors of the mid-latitudes, the Macquarie Island record
suggests this has also been accompanied by increasing variability (Fig. 5d).
To explore whether this is manifested across the wider Pacific we compared
our 140-year temperature reconstruction to key datasets (Fig. 9). Parallel
changes in SST magnitude and trend in the southwest Pacific using both the
LOVECLIM model output and HadISST (Rayner et al., 2003) is consistent with
our reconstruction of subantarctic island temperatures (Figs. 3, 8, and 9).
Intriguingly, the inferred increasing westerly winds and warming Southern
Ocean in the southwest Pacific have been accompanied by a regional order of
magnitude decline in marine vertebrate populations (McMahon et al., 2005; Morrison et al., 2015),
suggesting that the increased interannual temperature variability may have played
a role, and this will form a focus for future work. Importantly, we find a
comparable increase in temperature and variance in the Nino 3 region,
supporting our contention that the tropics are a major driver of variability
across the subantarctic Pacific and implying similar variability may be
expressed across other sectors of the Southern Ocean, albeit lagged by 1–3
months (Fig. S17). To test this we utilise snow core accumulation records
from coastal West Antarctica, a region identified as sensitive to atmospheric
pressure anomalies associated with the PSA (Thomas et al., 2008) (Fig. 6).
Previous studies have reported a mid- to late twentieth century increase in
precipitation associated with a deepening of the Amundsen Sea Low (ASL)
(Thomas et al., 2008, 2015), where strong northerlies advect
warm South Pacific air masses over the continent, resulting in
orography-driven precipitation over the southern Antarctic Peninsula (Gomez
ice core; Fig. 9f) and the West Antarctic coastal sites Bryan Coast and
Ferrigno (Fig. S18). Importantly, the observed twentieth century increase
appears to be confined to the Antarctic Peninsula and West Antarctic coast,
with the magnitude decreasing from east (Gomez) to west (Ferrigno); in marked
contrast, the observed increase is not recorded in the continental interior
(Thomas et al., 2015). Whilst the ASL is generally considered
quasi-stationary because of the large number of low-pressure systems in this
sector of the circumpolar trough (Hosking et al., 2013), the snow-core-derived increases in precipitation are accompanied by an increase in 30-year
running mean of the standard deviation, suggesting increased variability in
the ASL region that is unusual in the context of the past 300 years, with the
Gomez site most sensitive to changes in synoptic conditions.
Whilst we cannot preclude that the climate teleconnections may have been
different prior to the 1940s, the parallel changes in variance observed
across the Pacific suggests this is not likely (Fig. 9). This interpretation
is supported by the recently reported stepped increase in spring–summer
rainfall over the South Atlantic during the 1940s, a shift apparently
unprecedented over at least the last 6000 years, and interpreted to be a
consequence of highly seasonal changes in atmospheric pressure over the
Amundsen–Bellingshausen seas (Turney et al., 2016a). Although analysis of the
most recent decade suggests a weakening of the PSA (Trenberth et al., 2014),
the observed persistently high spring–summer Pacific variance and increase in
Atlantic precipitation (Turney et al., 2016a) suggests that Rossby wave
penetration of the high latitudes remains substantial when placed in the
context of the last 140 years (Fig. 9).