Antarctic Bottom Water Variability in a Coupled Climate …web.maths.unsw.edu.au/~asantoso/aabwms_jpo.pdf · Antarctic Bottom Water Variability in a Coupled Climate Model ... The

Antarctic Bottom Water Variability in a Coupled

Climate Model

Agus Santoso∗ and Matthew H. England

Climate Change Research Centre, University of New South Wales, Sydney, New South Wales,

Australia

October 15, 2007

Journal of Physical Oceanography (revised)

∗Corresponding author address : Agus Santoso, Climate Change Research Centre, University of New

South Wales, Sydney, NSW 2052, Australia. E-mail: [email protected]

Abstract

The natural variability of the Weddell Sea variety of Antarctic Bottom Water (AABW)

is examined in a long-term integration of a coupled climate model. Examination of passive

tracer concentrations suggests the model AABW is predominantly sourced in the Weddell

Sea. The maximum rate of the Atlantic sector Antarctic overturning (ψatl) is shown to effec-

tively represent the outflow of Weddell Sea deep and bottom waters and the compensating

inflow of Warm Deep Water (WDW). The variability of ψatl is found to be driven by surface

density variability which is in turn controlled by sea surface salinity (SSS). This suggests

that SSS is a better proxy than SST for post-Holocene paleoclimate reconstructions of the

AABW overturning rate. Heat-salt budget and composite analyses reveal that during years

of high Weddell Sea salinity, there is an increased removal of summertime sea ice by en-

hanced wind-driven ice drift, resulting in increased solar radiation absorbed into the ocean.

The larger ice-free region in summer then leads to enhanced air-sea heat loss, more rapid

ice growth, and therefore greater brine rejection during winter. Together with a negative-

feedback mechanism involving anomalous WDW inflow and sea-ice melting, this results in

positively correlated θ − S anomalies that in turn drive anomalous convection, impacting

on AABW variability. Analysis of the propagation of θ − S anomalies is conducted along

an isopycnal surface marking the separation boundary between AABW and the overlying

Circumpolar Deep Water. Empirical orthogonal function analyses reveal propagation of

θ − S anomalies from the Weddell Sea into the Atlantic interior with the dominant modes

characterised by fluctuations on interannual to centennial time scales. While salinity vari-

ability is dominated by along-isopycnal propagation, θ variability is dominated by isopycnal

heaving, which infers propagation of density anomalies with the speed of baroclinic waves.

1

1. Introduction

Antarctic Bottom Water (AABW) forms a major component of the global ocean ther-

mohaline circulation. Originating around the periphery of Antarctica, AABW mixes with

more saline and warmer Circumpolar Deep Water (CDW) as it spreads into the abyssal

basins of the world ocean (Mantyla and Reid 1983; Jacobs 2004). In the Atlantic sector,

AABW further mixes with lighter water masses as it flows equatorward reaching the North

Atlantic where interaction with North Atlantic Deep Water (NADW) occurs (see, e.g., Brix

and Gerdes 2003). AABW variability influences the stability of the global overturning cir-

culation and thus exerts an influence on the Earth’s climate over long time scales. However,

a better understanding of the spatial and temporal characteristics of AABW variability, es-

pecially on timescales beyond decades, is hampered by a lack of any extended observational

record. Furthermore, at present, it is not feasible to directly measure the AABW overturn-

ing variability (see also Latif et al. 2004). The present study aims to provide insight into

the evolution of AABW overturning and property anomalies on interannual to centennial

time scales operating in a coupled climate model, completing a series of papers exploring

the natural variability of Southern Ocean water masses (Rintoul and England 2002, San-

toso and England 2004, Santoso et al. 2006). Here a focus is placed on the Atlantic sector

AABW and the Weddell Sea variety of deep and bottom waters.

Sources of AABW include both the contribution from shelf waters at several sites around

the Antarctic continental margin (Baines and Condie 1998) and the upwelling of CDW south

of the Antarctic Circumpolar Current (ACC; see a review by Orsi et al. 1999). Despite

significant input from the Ross Sea and other regions such as the Adelie land region (Rintoul

1998; Orsi et al. 1999), the Weddell Sea is regarded as the most prominent and active region

2

for bottom water formation (e.g., Carmack 1977; Orsi et al. 1999). The process of AABW

formation has been described in several previous studies (e.g., Foster and Carmack 1976;

Baines and Condie 1998; Rintoul 1998; Orsi et al. 1999; Meredith et al. 2000; Foldvik

et al. 2004); here we briefly describe the Weddell Sea variety. Upwelled CDW enters the

Weddell Sea where mixing with the overlying colder and fresher winter water occurs. Winter

water exists year-round in the surface mixed layer as a remnant of the cold layer produced

during sea-ice formation (Foster and Carmack 1976). This modified CDW, now referred to

as Warm Deep Water (WDW; Whitworth and Nowlin 1987), mixes further with the high

salinity southwestern Weddell Sea Shelf Water (−1.9◦C, 34.7) to form Weddell Sea Bottom

Water (WSBW; −1.3◦C, 34.65). WSBW then mixes with the less dense CDW as it flows

down the continental shelf to form Weddell Sea Deep Water (WSDW), gaining buoyancy

to flow over sills as AABW (typical θ − S of ≈ −0.4◦C, 34.66; Whitworth and Nowlin

1987). Nevertheless, the major ingredient of AABW is CDW, as suggested by the analysis

of Foster and Carmack (1976) who showed that AABW sourced from the Weddell Sea is

composed of ≈ 62.5% CDW, 25% shelf water, and 12.5% winter water (see also Whitworth

et al. 1998).

Capturing AABW properties, formation, and pathways in general circulation models

(GCM) is a challenging task, requiring a realistic representation of shelf processes (includ-

ing Antarctic sea ice and ice shelves), downslope flows, and convective overturning (Goosse

et al. 2001; Doney and Hecht 2002; Stossel et al. 2002). To date no climate model has been

capable of simulating the correct properties of the abyssal oceans. However, considerable

improvements have been achieved since the early model developments of Bryan (1969) and

Cox (1984). For example, one of the most common deficiencies in climate models is that

3

the deep ocean is not sufficiently dense, with deep and bottom waters too fresh. While

Antarctic wintertime surface salinity adjustments can yield a better AABW representation

in GCMs (e.g., England 1993), they generally equate to spurious air-sea freshwater fluxes

(Toggweiler and Samuels 1995). Furthermore, the long ventilation time scales of AABW re-

quires a multi-century integration of a fully coupled GCM, currently only possible at coarse

resolution. Explicit representation of key processes such as convection and bottom bound-

ary currents is beyond the present-day class of models used to predict anthropogenic climate

change (some progress in the parameterisation of bottom boundary layers has improved the

representation of downslope flows; Doney and Hecht 2002). However, climate-scale models

generally capture realistic net production rates of AABW (England et al. 2007; manuscript

in preparation for J. Phys. Oceanogr.), and in some cases reasonable T − S properties and

CFC uptake (e.g., Doney and Hecht 2002). In such cases they can provide a meaningful

way to examine the physics of long-term natural variability of AABW over interannual to

centennial time scales.

There have been relatively few studies of the observed variability of AABW properties

and its ingredients on seasonal to decadal time scales. Coles et al. (1996) found AABW

cooling and freshening of 0.05◦C and 0.008 psu along constant density surfaces in the

Argentine Basin over the period 1980−1989, accompanied by observable warming at abyssal

depths. They suggested this θ − S change would be linked to convective events in the

Weddell Sea. Hogg and Zenk (1997) documented warming in the bottom waters of the

Vema Channel of about 0.03◦C, accompanied by a decrease in the northward bottom water

transport. They proposed that these changes are a response to a reduction in bottom water

production. This warming appears to have continued until 2005 as found by Johnson and

4

Doney (2006) in the Brazil, South Georgia, and Argentine Basins. Warming of WDW and

WSDW by 0.1◦ − 0.2◦C and 0.05◦C, respectively, were observed by Meredith et al. (2001)

in the eastern Scotia Sea between 1995 and 1999. Changes in the WSDW properties at

the shelves of the Weddell Sea, and changes in the wind-driven gyre, were listed as possible

causes for the warming. More recently, Fahrbach et al. (2004) documented θ−S fluctuations

in the Weddell Sea over 1990 to 2002. A warming trend of WDW is observed from 1992

to 1998 at the prime meridian which is consistent to that documented by Robertson et al.

(2002). The warming trend is then followed by a cooling trend. Changes in the θ − S of

WSDW and WSBW are also documented by Fahrbach et al. (2004) with an amplitude

of the order of 0.01◦ − 0.02◦C, 0.001−0.002 psu, respectively. They proposed that these

changes are caused by variations in atmospheric circulation in response to climate modes

such as the Antarctic Circumpolar Wave (ACW) and the Southern Annular Mode (SAM),

which can impact the inflow of ACC waters into the Weddell Sea.

Observational studies of low frequency AABW variability up to centennial time scales

are naturally absent given the short measurement record available. Modelling studies inves-

tigating AABW variability on interannual-decadal time scales have recently emerged, such

as those by Stossel and Kim (1998; 2001) using a coupled sea ice-ocean GCM. Stossel and

Kim (1998) found a 4-yr oscillation in AABW outflow and ACC transport confined within

the Weddell Sea-Drake Passage region, generated internally by the sea ice-ocean system.

By switching the wind forcing from monthly climatological to daily values, Stossel and Kim

(2001) found a decadal mode associated with enhanced convection in the southern Weddell

Sea, which was suggested to be induced by entrainment of anomalous CDW. The Stossel

and Kim studies above imply the importance of an active dynamic-thermodynamic sea-

5

ice model for a meaningful investigation of high-latitude variability (see also Stossel et al.

1998). In this study, we extend the investigation of AABW variability and its mechanisms

to a fully coupled global atmosphere-ice-ocean model integrated over multi-millennial time

scales.

The strength of the overturning cell emanating in the polar region of the Southern Ocean

(far left-hand cell of Fig. 3a) is widely used in modelling studies to represent the strength

of AABW production (e.g., Drijfhout et al. 1996; Brix and Gerdes 2003). This cell strength

is difficult to estimate from observations (Latif et al. 2004). Instead, volumetric analyses

based on chlorofluorocarbon and mass budgets are generally used (Orsi et al. 1999). In

this study we will simply analyse variability in AABW formation rates via variability in the

polar meridional overturning cell. We will also assess variability in AABW θ−S properties

and how this relates to variability in production rates, and atmosphere-ice-ocean surface

property fluxes.

The purpose of this paper is to provide an extensive analysis of AABW overturning and

θ − S variability on interannual to centennial time scales in a long-term integration of a

coupled climate model. Of particular interest is how the variability in AABW overturning

and properties is influenced by surface θ−S conditions. The coupled model and its bottom

water features are described in section 2. Section 3 investigates the link between AABW

overturning and surface properties. The mechanisms of variability are investigated in section

4. In section 5, we assess the propagation of θ− S anomalies into the interior. Finally, the

study is summarised in section 6.

6

2. AABW in the climate model

a. The climate model

The model used in this study is the CSIRO Mark 2 10,000-yr integrated natural pre-

industrial CO2 coupled ocean-atmosphere-ice-land surface model. A full description of the

model can be found in Gordon and O’Farrell (1997) and Hirst et al. (2000), here we only

summarise briefly. We analyse 1000 yr of model data from the latter stages of the 10,000-yr

run, by which time the model exhibits very minimal drift.

The atmospheric model is discretised on nine levels in a sigma coordinate system. Pa-

rameterisation of land surface interactions follow the soil-canopy model of Kowalczyk et

al. (1994). The sea-ice model includes the cavitating fluid rheology of Flato and Hibler

(1990), ice thermodynamics (Semtner 1976) and sea-ice dynamics allowing advection and

divergence of sea ice by wind stress and ocean currents (see O’Farrell 1998 for details).

The ocean model is based on the Bryan-Cox code (Cox 1984) with horizontal resolu-

tion ≈ 5.6◦ longitude × 3.2 latitude, matching that of the atmospheric component. In

the vertical, the model has 21 levels of irregular grid box thickness. The model captures

major land-masses and bottom bathymetric features; although due to coarse resolution,

topographic features are broader than observed. The Gent-McWilliams parameterisation

(GM; Gent and McWilliams 1990; Gent et al. 1995) is implemented with horizontal back-

ground diffusivity set at zero. Along isopycnal mixing of Cox (1987) and Redi (1982) is

implemented with an isopycnal tracer diffusivity of 1× 107 cm2 s−1. Convective overturn is

simulated by applying an enhanced vertical diffusivity in regions of static instability. Con-

stant annual but seasonally varying air-sea flux adjustments (heat, freshwater, and wind

stress) are included in the coupling between the ocean and atmosphere (and ocean and sea

7

ice) to reduce long-term climate drift. There are no flux corrections applied between the at-

mosphere and sea-ice components. The root mean square of the flux adjustment terms over

the Southern Ocean south of 50◦S is 33.5 W m−2 for heat and 0.69 m yr−1 for freshwater

(Hirst et al. 2000). There is no Newtonian damping component to these flux adjustment

terms.

The implementation of the GM eddy-induced mixing parameterisation allows the elimi-

nation of background horizontal diffusivity in the model. This is worth mentioning as it has

an important implication on AABW formation in the model. The inclusion of GM results in

1) a better subsurface stratification, 2) colder and more saline (and therefore denser) deep

waters, due to the flattening of isopycnals, and 3) denser downslope flows due to a lack of

erosion by unrealistic horizontal diffusive fluxes (e.g., Hirst and McDougall 1996). The GM

parameterisation thus results in much reduced open-ocean convection, especially at high

southern latitudes (Hirst et al. 2000). This effect is particularly desirable as more AABW

tends to form via near-boundary convection adjacent to the Antarctic coast, exhibiting a

closer correspondence to the real system. Furthermore, the reduction of Southern Ocean

spurious convection in turn allows for smaller flux adjustment terms in the region. A com-

parison between this simulation and a more recent version without flux adjustment shows

no obvious influence of the flux adjustment terms on the model’s climate variability (Hunt

2004). The favourable effects of GM on ocean water-masses together with the fully coupled

nature of the model, and its efficient computational cost for multi-millenial integrations,

allow us to implement the model for investigating large-scale AABW variability and its

mechanisms on long time scales.

8

b. Modelled AABW

The spreading and ventilation pathways of AABW are illustrated in Fig. 1. This

diagram shows passive tracer concentration at the bottom-most cells at 50 and 150 yr after

release of tracer at the surface where it is set to 100% (after O’Farrell 2002). The bottom

current velocities and bathymetry contours are also shown in Fig. 1. The tracer is at

highest concentrations in the Weddell Sea, with a second weaker signal originating in the

Ross Sea. Since the model AABW is predominantly produced in the Weddell Sea, we will

focus our analyses on the Weddell Sea variety of AABW.

Generally speaking the model’s advective time scales will be slower than observed, and

the pathways of AABW ventilation will be broader compared to the real system. Nonethe-

less, as in observations, the tracer from the Weddell Sea flows north westward and eastward,

spreading into the Argentine Basin and the Weddell-Enderby Plain respectively (Orsi et

al. 1999). The overflow of tracer into the Scotia Sea reaching the Drake Passage is also

consistent with observations of WSDW (Naveira Garabato et al. 2002). However, it may

be noted that there is excessive intrusion of tracer from Cape Basin into the Angola Basin

in the eastern Atlantic, in contrast to the real ocean where there is thought to be virtually

no or only little bottom water of southern origin found in the Angola Basin (Reid 1989;

Larque et al. 1997). This discrepancy is likely due to the unresolved obstruction of the

Walvis Ridge in the model. Furthermore, the tracer concentration minimum in the Brazil

Basin is likely due to the unresolved Vema and Hunter Channels, which would otherwise

allow inflow of WSDW from the Argentine Basin (Larque et al. 1997; Hogg et al. 1999).

Eastward spreading of WSBW into the Indian Ocean is apparent with a decreasing tracer

concentration due to mixing with the overlying CDW. The bottom flow from the Weddell

9

Basin into the Mozambique and Crozet Basins is consistent with the findings of Mantyla

and Reid (1995) and Haine et al. (1998). While the model deep and bottom waters are

not as saline or dense as in the real ocean (Hirst et al. 2000), the model overall reproduces

key features of Weddell Sea deep and bottom water ventilation rates and pathways. This

aspect of the model coupled with its inexpensive computation is what makes it relevant

for studying the mechanisms driving AABW variability on long time scales, and how this

variability is transferred into the ocean interior.

The positive velocities at 68◦S within the Weddell Sea shown in Fig. 2 indicate WSDW/WSBW

as the origin of AABW in the Atlantic sector. These water masses are contained under-

neath an isopycnal surface (labelled as σ41.50; solid contour in Fig. 2 and Fig. 3b) sep-

arating AABW from CDW. This isopycnal surface captures waters with relatively high

tracer concentration as far north as the equator, thus satisfying the definition of AABW;

namely waters sourced from the Antarctic surface that are eventually found in the abyssal

oceanic basins. Note that σ41.50 is a ‘patched’ potential density surface corresponding to

ρ3 = 1041.50 kg m−3, locally referenced over five pressure levels; namely, 0, 1000, 2000,

3000, and 4000 db (see Reid 1994 for details on the construction of patched density sur-

faces). A deeper potential density surface corresponding to ρ4 = 1045.95 kg m−3 (referenced

to 4000 db; σ45.95 surface hereafter) is also shown in Fig. 2 (dashed contour) and Fig. 3b.

The outflow of WSDW/WSBW is compensated by the inflow of WDW which is embod-

ied within the region of negative velocities above σ41.50 in the Weddell Sea (Fig. 2). This

inflow-outflow regime constitutes the Antarctic meridional overturning circulation (MOC)

in the Atlantic sector shown in Fig. 3b. The global meridional overturning in the model is

shown in Fig. 3a, depicting the meridional cells of the world ocean in a zonally-integrated

10

perspective. The strength of the negative Antarctic overturning cell south of 60◦S is com-

monly taken to be the AABW formation rate in ocean GCMs (labelled in Fig. 3a). In the

model this overturning cell represents a maximum transport of up to 10.5 Sv with a mean

of 8.5 Sv; at the lower end of the observed range of 5−15 Sv (e.g., Gill 1973; Carmack

1977; Jacobs et al. 1985; Orsi et al. 1999). The bottom-water layer below σ41.50 shown

in Fig. 3b captures the Atlantic sector of the Antarctic cell and the lower portion of the

Atlantic sector abyssal cell to the north. It may be noted that the northward penetration

of AABW into the North Atlantic, as suggested by the abyssal cell in Fig. 3b, is too far

north in the model due to the weak formation of lower North Atlantic Deep Water − a

common problem in ocean GCMs (England and Holloway 1998). The time-series of the

Atlantic sector overturning is compared to that of the global Antarctic overturning cell in

Fig. 3c, demonstrating that AABW in the model is predominantly sourced in the Weddell

Sea. The variance of the Atlantic sector Antarctic overturning accounts for about 70% of

the total variability of global AABW production.

Figure 4 demonstrates that waters on σ45.95 are generally more rapidly ventilated, colder,

and more saline than those on σ41.50. Specifically, Fig. 4a shows a scatter plot of tracer

concentration along σ41.50 and along the bottom-most cells against the tracer concentration

along σ45.95 in the Atlantic sector. Relative to tracer concentrations on σ45.95, σ41.50 contains

lower concentration when the tracer concentration on σ45.95 is lower than 75%. The two

layers exhibit comparable concentrations when the concentration is higher than 75%. The

opposite holds for the bottom-most cells, with σ45.95 less ventilated as it overlies the bottom

grid cells. This suggests that the model σ41.50 isopycnal captures the WSDW while σ45.95

marks the WSBW layer, both of which are well ventilated adjacent to Antarctica. Away

11

from the Antarctic margin, the ventilation on σ41.50 decreases while the bottom layers are

more rapidly flushed by the northward spreading Weddell Sea bottom waters which are

cold and saline (Fig. 4b, c).

3. Variability of Weddell Sea Bottom Water

a. A meridional overturning representation of outflow and sinking

In this section, we demonstrate that the maximum rates of the Atlantic sector Antarctic

overturning (see Fig. 3; denoted ψatl hereafter) can be adopted to conveniently represent

sinking and outflow rates of the Weddell Sea bottom waters in the model. In this case,

examining the variability of sinking and outflow rates of Weddell Sea deep and bottom

waters is equivalent to examining the variability of ψatl.

The outflow of WSDW/WSBW can be represented by the integral of meridional veloc-

ities (v) within the sector 65◦W−3◦E at 68◦S (Fig. 2) calculated as:

Vout =∫

3◦E

65◦W

∫ zσ

bottom

vR cos(68◦S) dz dϕ (1)

where zσ is the mean depth of σ41.50 below 1500 m, R the radius of the Earth, dz the

thickness of grid box at a model depth level and dϕ the longitudinal width of the model

grid. The integral in Eq. (1) purposefully includes both northward and southward velocities

to ensure that any local recirculations are not added to the net outflow diagnostic. Similarly,

the inflow of WDW is calculated as the integral of meridional velocities from the surface to

above the depth of σ41.50:

Vin =∫

3◦E

65◦W

∫0

zσ

vR cos(68◦S) dz dϕ. (2)

12

The integral of downward velocities south of 68◦S is also computed and is maximum at

1500-m depth where ψatl is most rapid. Table 1 summarises the mean values and ranges

of the above integrated transports as well as their correlation coefficients against ψatl. The

average value of the above integral transports is 8.0 Sv with a range of ≈ 4 Sv. The

correlations in Table 1 suggest that high ψatl corresponds to high sinking and outflow of

Weddell Sea deep and bottom waters leading to high inflow of WDW. These correlations

of near unity over a 1000-yr time series imply that ψatl is an accurate estimate of Weddell

Sea deep and bottom water transports in the model. For the rest of this study, we use ψatl

as a direct measure of WSDW/WSBW formation and outflow in the model. While this is

certainly valid for the Atlantic sector, the global meridional overturning may not necessarily

be a good estimate of the total production of AABW (England et al. 2007; manuscript in

preparation for Journal of Phys. Oceanogr.). A simple but accurate diagnosis of AABW

outflow, both in models and observations, remains a topic of ongoing research.

The MOC is often also viewed in the density-latitude plane. The model’s Atlantic sec-

tor MOC in density coordinates was calculated for the full 1000-yr period, and compared

to the latitude-depth ψatl diagnostic. As demonstrated in Fig. 3d for an arbitrary 200-yr

record, the time series of AABW formation in density coordinates (ψatl|σ) is highly signifi-

cantly correlated to the z -level overturning (ψatl), with a correlation coefficient (r) of 0.86,

in which ψatl leads ψatl|σ by 1 yr. However, the ψatl|σ metric is more weakly correlated to

the meridional transport metrics of Eqs. (1), (2) (r ≈ 0.6) than when using ψatl (r ≈ 0.9).

We thus find it more suitable to employ the z -level overturning diagnostic of bottom water

production in this study.

13

b. Overturning variability linked to surface buoyancy

The extent to which variability in surface buoyancy impacts the rates of deep and

bottom water outflow and formation is assessed here. Figure 5 shows lagged-correlation

maps in which surface density leads ψatl by 2 and 4 yr, as well as the zero-yr lag analysis.

Significant correlations occur largely in the Weddell Sea, with some additional regions of

high correlation to the west of the Antarctic Peninsula. The highest correlation (r = 0.72)

is seen in the central Weddell Sea when surface density leads ψatl by about 3 years. This

equates to an increase in surface density being followed by an enhanced overturning a few

years later. Taking a spatial average over the region indicated in Fig. 5, the time series

of the sea surface salinity and temperature (denoted as SSSwed and SSTwed hereafter)

are plotted against ψatl in Fig. 6. It is apparent that variations in ψatl are accompanied

by fluctuations in SSSwed and SSTwed in such a way that anomalously saline and warm

surface waters lead to a more vigorous overturning of deep and bottom waters. The positive

correlation between the overturning and each of SSS, SST, and surface density implies that

SSS controls density, and thus overturning, as SST would be negatively correlated to both

overturning and density if it were the driving component.

The power spectrum of ψatl is presented in Fig. 7a showing spectral peaks at periods

over interannual to centennial time scales. It can be seen in Fig. 7b that SSS also exhibits

spectral peaks similar to those of ψatl on decadal to centennial time scales. On the other

hand, the only spectral peaks of SST that coincide with those of SSS and ψatl are the 21-yr

and 32-yr periodicities (Fig. 7c). Furthermore, it is noted that the correlations for ψatl are

higher against SSSwed (r = 0.78) than SSTwed (r = 0.52) with ψatl lagging surface θ − S

by 3 years. This suggests that SSS is a better proxy for the AABW overturning cell for

14

paleoclimatic reconstructions.

The surface density perturbation signals are transmitted to the bottom depth as il-

lustrated in Fig. 8a, showing lagged correlations between ψatl and the spatially-averaged

density from the surface to the deepest model level. A similar correlation analysis for θ−S

(Fig. 8b, c) reveals positive (negative) θ − S anomalies created at the surface a few years

prior to an enhanced (weakened) overturning. While salinity is positively correlated to ψatl

at all depth levels (Fig. 8b), θ is negatively correlated below the mixed layer (Fig. 8c). This

is because cooling (warming) in the interior is a result of enhanced (reduced) convection,

advecting more (less) cold water from the surface layer into the interior − in response to the

surface density increase (decrease). The vertical transmission of surface density and salinity

anomalies into the interior is confirmed by Fig. 8d-e, showing lagged correlations between

density and salinity at the surface and those at depth. However, the correlation pattern

for temperature (Fig. 8f) is in contrast to that shown in Fig. 8c, because the associated

convective overturning events are driven by surface salinity anomalies, not temperature

anomalies, thus resulting in a stronger subsurface-interior connection.

The θ− S anomaly distributions at various depths in connection with overturning vari-

ability is depicted in Fig. 9. Here we separate the θ−S anomalies that correspond to years

of anomalously high ψatl from those occuring during anomalously low ψatl, at the time lags

at which the correlation of density and ψatl is at a maximum. Years of anomalously high

and low ψatl are defined as those when ψatl anomalies exceed one standard deviation above

and below the long-term mean. These θ− S anomalies are presented in αθ′

− βS′

space in

Fig. 9, so that the dominance of θ vs S in controlling the density variations can be assessed.

A similar analysis can be found in Santoso et al. (2006; see their Appendix). However, here

15

we consider deviations from the 1000-yr mean as denoted by θ′

, S′

, in contrast to Santoso

et al. (2006) who considered year-to-year changes (θt, St). It can be seen that the θ − S

anomaly distributions during low ψatl years are almost the mirror image of the anomalies

for high ψatl years. Thus, we limit the following discussion to the case of high overturning

anomalies.

At the surface (Fig. 9a), the density anomalies are distributed in the warming-salination

regime below the absolute density-compensation line where Rρ = 1. The average of the

density ratios, Rρ = αθ′

βS′ , is ≈ 0.1. This implies the dominance of surface salinity increase

over warming on the positive density perturbations that ultimately set vigorous overturn-

ing. In the interior, such as at 410, 2125, and 4375 m (Fig. 9b, c, d), the distribution of

density anomalies leaks into the cooling regime (below Rρ = 0). This illustrates the effect

of convective adjustment as described above. It may be noted that the anomaly distribu-

tion becomes more dispersed with increasing depth, intruding into the freshening regime

occassionally. However, in general, the composite anomalies shown in Fig. 9 suggest that

years of high bottom water formation and outflow are initiated by salination and warming

of surface waters, which then leads to anomalous convection, resulting in cooler and saltier

bottom waters at fixed depths. The opposite holds during years of low bottom water pro-

duction. The composites of θ − S variations at constant density levels do not show such

coherent patterns (Fig. not shown), likely due to spatial aliasing (refer to section 5 for

description of the spatial patterns of the θ − S anomalies).

16

4. Mechanisms of variability

It was demonstrated in section 3b that fluctuations in Weddell Sea Bottom Water pro-

duction are associated with salinity-driven surface density perturbations accompanied by

positively correlated SST and SSS anomalies (see also Fig. 6a, b). To reveal the mech-

anisms involved in setting these variations in SSS and SST, surface heat and salt budget

analyses are conducted. First note that the freshwater flux into the ocean in the coupled

model can be seen as an equivalent negative salt flux into the ocean. For brevity, this

‘equivalent salt flux’ is hereafter simply referred to as the ‘salt flux’. Comparisons of the

standard deviations of the budget terms spatially-averaged over the Weddell Sea (see the

boxed region in Fig. 5) are presented in Tables 2 and 3. Net surface heat and salt fluxes

dominate the annual-mean SST and SSS variability. Short-wave radiation (Qsolar) is found

to dominate the surface heat flux variability, while the surface freshwater flux is dominated

by variations in the sea-ice meltwater rate. This dominance of Qsolar and sea-ice meltwater

is robust at all time scales, as confirmed by heat−salt budget analyses on data filtered

with various band-pass frequencies (not shown). Our analysis suggests variations in sea-ice

coverage plays a crucial role in regulating Weddell Sea surface water density, via its direct

link to variations in solar heat flux and the sea-ice meltwater/brine rejection rate.

The composite means of sea-ice concentration, ice-ocean salt flux, and solar heat flux

for years of high salinity Weddell Sea surface water are shown in Fig. 10. The high salinity

years are those when the spatially-averaged salinity within the indicated region exceeds

one standard deviation above the long-term mean. Our focus here is on SSS as salinity,

not temperature, regulates surface density perturbations in the Weddell Sea bottom water

formation region (section 3b). Seasonal effects are shown by presenting the annual-mean,

17

summer (averaged over December−March), and winter (June−September) composites. The

composites for the low SSS anomaly years are not displayed as they are more or less the

mirror-image of the high salinity composites. Accordingly, our discussion will focus on the

results of the high salinity composite analysis.

Surprisingly the annual-mean anomalies in the region of interest are dominated by

summer-time variability (Fig. 10). This is supported by the observations of Zwally et

al. (2002), who documented larger long-term changes in sea-ice coverage in summer than

in winter, over 1979−1998. During years of high Weddell Sea surface salinity, summer ice

coverage is anomalously low (Fig. 10a), meltwater input is low1 (Fig. 10b), and incoming

solar radiation is anomalously high (Fig. 10c). Anomalies during winter are, in contrast,

generally weak within the Weddell Sea, apart from higher than average brine rejection in

the southwest region (Fig. 10b, right-hand panel). The anomalous pattern of short-wave

radiation absorbed by the ocean over summer (Fig. 10c, middle panel) coincides with that

of low sea-ice concentration, confirming the link between sea-ice coverage and surface heat

fluxes in regulating SST variability. The winter months, however, exhibit little anomalous

sea-ice concentration (Fig. 10a, right panel), yet they are accompanied by positive ice-ocean

salt flux anomalies (Fig. 10b, right panel). This counter intuitive result will be explained

here below.

Figure 11 presents the composites of the annually-averaged SSS, SST, wind stresses,

and surface net salt flux for both the high and low salinity composite means. Unusually

high surface salinity (Fig. 11a) and temperature (Fig. 11b) periods are accompanied by

1The mean ice-ocean equivalent salt flux in summer is negative (i.e., due to melting of sea ice), except

over a small region in the eastern Weddell Sea, so the positive ice-ocean salt flux anomaly in Fig. 10b

implies anomalously low meltwater input.

18

strengthened westerlies (Fig. 11c) and anomalous katabatic winds (Fig. 11d) which would

drive sea ice northward in the western flank of the Weddell Gyre (Fig. 10a, left panel;

see also Uotila et al. 2000; Harms et al. 2001). This explains the reduction in sea-ice

concentration in the west during summer (Fig. 10a, middle panel) and the increase in sea

ice in winter to the north (Fig. 10a, right panel). Consequently, the removal of sea ice over

the region reduces the amount of ice available for melting in summer, thus explaining the

low meltwater flux shown in Fig. 10b (middle panel). The overall reduction in the summer

sea-ice coverage allows enhanced ocean cooling by latent and sensible heat fluxes (Fig.

not shown), although the warming by enhanced incoming solar radiation still dominates.

Approaching winter, enhanced atmospheric cooling over the larger than normal ice-free area

leads to higher than average brine rejection (Fig. 10b), with the ice concentration itself

merely recovering to normal wintertime levels (Fig. 10a, right panel). This indicates that

a negative ice-ocean feedback loop is limiting wintertime ice anomalies despite substantial

variations in the summertime ice coverage and seasonal ice-ocean salt fluxes. The ice-ocean

salt flux composite mean dominates the net air/ice-ocean salt flux anomaly (Fig. 11e) as

demonstrated in Table 3. The mechanism described above is illustrated by the schematic

diagram shown in Fig. 14.

It is interesting to note that an annular pattern appears in the zonal wind stress com-

posites of high and low salinity years (Fig. 11c), suggesting an influence of the Southern

Annular Mode in forcing SST and SSS variations in the Weddell Sea. An EOF analysis

conducted on the annually-averaged zonal wind stress (τ x) exhibits a zonally-symmetric

pattern as its dominant mode, accounting for 26% of the total τ x variance (Fig. 12a). The

corresponding principal component (PC) time series extracted from the analysis (Fig. 12b)

19

is correlated to SSSwed and SSTwed with only a modest correlation coefficient of ≈ 0.27,

yet with a 1000-yr time series this is well above the 99% confidence level, with the PC time

series leading by 1 year. The power spectrum of the PC time series exhibits significant

signals above the background noise with peak periods of ≈ 8 and 30 years (Fig. 12c),

thus contributing to the interannual to interdecadal variations in SSSwed and SSTwed. It is

likely that other modes of climate variability, such as the model’s ACW and Pacific-South

America (PSA) modes, may also influence the variability, however, further investigation

on this topic is beyond the scope of the present study. What is apparent from the above

analyses is that sea-ice variability plays a key role in Weddell Sea salt and heat content

variations, which ultimately drive fluctuations in bottom water formation and outflow rates

in the Atlantic sector of the model.

While it is evident that direct atmospheric forcing controls variations in SSS and SST via

sea-ice variability, there is also evidence of an internal feedback mechanism that modulates

the quasi periodicities observed in ψatl and SSS (refer to Fig. 7). A lagged correlation

analysis presented in Fig. 13 explains a negative feedback mechanism involving sea ice,

SSS, and the Antarctic overturning (ψatl), as implied by the opposite signs of the correlation

coefficients at negative and positive time lags. The correlation analysis is presented based on

the raw and band-pass filtered time series to isolate interdecadal signals with periods in the

range 10 to 50 years. It can be seen that the negative feedback becomes more apparent with

the time series filtered to retain only signals of 10−33-yr periodicities. Positive correlations

at the positive time lags indicate the ice-to-ocean salt flux (Hice) leads an increase in ψatl

by about 5 years (Fig. 13a) via higher surface salinity (Fig. 13b), which in turn generates

higher density surface waters (section 3b), leading to enhanced ψatl (Fig. 13c). The negative

20

correlations at negative time lags in Fig. 13a imply that an increase in ψatl (see schematic

diagram in Fig. 14b) leads to a reduction in Hice (or equivalently an increase in meltwater;

Fig. 14d) with a time lag of about 2 yr. This is because an increase in overturning causes an

increased inflow of WDW (see Table 1; Fig. 14b) which provides heat from the subsurface to

melt sea ice, thus lower SSS (Fig. 14d). This leads to a weakening of overturning which then

reduces the amount of heat injected under the sea ice, creating a higher salinity anomaly

at the surface (Fig. 14b). This negative-feedback cycle continues, linking variations in sea

ice and overturning via sea surface salinity variability.

5. Propagation of θ − S anomalies

We have explored how the Atlantic sector of the Antarctic overturning (ψatl) fluctuates

with surface properties in the Weddell Sea (section 3b) and the mechanisms that give rise

to this variability (section 4). We now shift our attention to the patterns and propagation

of θ − S anomalies into the abyssal Atlantic Ocean. For this purpose, we conduct analyses

on σ41.50 as this density surface extends from the Weddell Sea source region to the Atlantic

equatorial region. A standard deviation analysis of θ − S along σ41.50 reveals the largest

magnitude variability of ≈0.15◦C, 0.01 psu at the Weddell Sea outflow region (Fig. 15).

The magnitude of variability is considerably reduced northward into the abyssal Atlantic

and eastward into the Indian Ocean. A complex empirical orthogonal function (CEOF)

analysis is conducted to extract the spatial and temporal characteristics of various modes

of θ − S variability along σ41.50. Since θ − S vary coherently along isopycnal surfaces, it

is sufficient to just present the CEOF analysis of salinity on σ41.50. The propagation of

θ−S anomalies are depicted in the CEOF maps in Fig. 16, which present the three leading

21

CEOF modes accounting for 76.5% of the total S variance. These modes are well separated

according to the North rule (North et al. 1982). The temporal characteristics of the modes

are represented by the principal component (PC) time series shown in Fig. 17 together with

their power spectra. For comparison, the CEOF temporal characteristics of S variability

on a deeper isopycnal surface (σ45.95; equivalent to σ3 ≈ 41.54 kg m−3) are also shown in

Fig. 17.

The CEOF analysis (Fig. 16) shows that θ − S anomalies emitted in the Weddell

Sea propagate eastward and then northward into the Atlantic Ocean. The CEOF spatial

patterns on σ45.95 show similar patterns to those in Fig. 16, thus are not shown. CEOF-1

shows a broad region of high variance extending deep into the interior, while the higher

CEOF modes capture more intense variance close to the surface, where decadal-interdecadal

signals are more energetic (Fig. 17). Due to the slow integrative effects of isopycnal mixing,

the low-frequency surface variability (Fig. 7) is best preserved as θ−S anomalies propagate

into the ocean interior, while higher modes are damped. This is more apparent for the near

centennial time-scale signals in Fig. 7a, b which are picked up by CEOF-1 (Fig. 17).

Inspecting Fig. 17 further, fluctuations on interannual to interdecadal time scales are more

prominent on σ45.95 than on σ41.50, where signals of centennial and longer timescales become

evident. Also apparent in the PC time series of Fig. 17 (left column) is the fact that the

magnitude of θ−S variability is larger on σ45.95. This is because the deeper bottom waters

are more rapidly ventilated than the upper deep waters (see section 2) where mixing with

CDW takes place. Indeed, the apparent ≈ 330-yr signal on σ41.50 matches the dominant

time scale of θ − S variability in the CDW layer (Santoso et al. 2006).

The advective time scale of the propagating θ−S anomalies along isopycnals is depicted

22

in the Hovmoller diagram presented across 62◦S on σ41.50 (Fig. 18a, b; see inset for the zonal

transect location). The space-time gradients of the anomalies suggest that the travelling

signals cover around 125◦ longitude over ≈ 50 yr. At 62◦S, this corresponds to a speed of

≈ 0.4 cm s−1, comparable to the ocean current speeds in the region. Similar diagrams are

also presented for the θ − S anomalies propagating along isobars (Fig. 18d, e), referenced

to the long-term averaged depths of the σ41.50 surface. The steep gradient of S anomalies

along the σ41.50 isopycnal (Fig. 18a) is also apparent in Fig. 18d, and the sign of the

along-isopycnal and along-isobar S anomalies are also in phase. This suggests that salinity

variability is dominated by along-isopycnal anomalies. However, this is not the case for θ

anomalies, which are related to the effect of isopycnal displacements (i.e., heave). This is

evident in the close resemblance between the θ anomaly patterns along isobars (Fig. 18e)

and the σ41.50 depth anomalies (hσ) shown in Fig. 18c, both in terms of their phase and

space-time gradients. This is further supported by the high mean correlation coefficient of

0.84 between θ|z and h|σ across 62◦S, with θ|z leading h|σ by ≈ 4 yr in the Weddell Sea

region.

The rising and deepening of the σ41.50 surface (Fig. 18c) illustrates density anomalies

which can be seen propagating eastward from 31◦W to 25◦E before intercepting a westward

wave propagation. Inspection of the gradients of the westward signals suggests the time

taken to travel 45◦ of longitude is about 5 yr. This implies a speed of up to ≈ 1.5 cm

s−1, which is roughly three times the theoretical speed of unforced baroclinic Rossby waves

given by c = βg′

H0/f2, where f is the Coriolis parameter, β the meridional derivative of

f , H0 the depth of the σ41.40 surface, and g′

is the reduced gravity calculated using the

difference between the average densities above and below σ41.50. A scaling analysis yields

23

typical values of the following parameters in the south Indian sector: g′

= 0.003 m s−2,

f = −1.3 × 10−4 s−1, β = 1.14 × 10−11 m−1 s−1, and H0 = 2500 m, which yields c ≈ 0.5

cm s−1 at 60◦ latitude, matching the background current velocity. As shown by Qiu et al.

(1997), fast baroclinic waves travelling higher than twice the theoretical speed are expected

be found in subpolar regions of the Southern Hemisphere. In contrast, the speed of the

eastward propagating hσ signals in the Atlantic sector is of the order of 0.5 cm s−1 which

is comparable to the mean velocity of the background current at that depth.

The northward propagation of θ − S anomalies into the abyssal Atlantic is further

masked by baroclinic wave propagation as shown by the Hovmoller diagrams of θ − S and

hσ anomalies along a meridional section crossing the Argentine Basin (Fig. 19; see inset

for the transect location). The signature of the fast baroclinic waves is apparent in the

patterns of the along-isobar anomalies (Fig. 19d, e). It is worth mentioning that the north-

ward propagation of the high frequency signals appears to shut down at about 50◦S. This is

because the anomalies join the ACC eastward as they flow to the north. Northward propa-

gation into the Argentine Basin mainly involves the low-frequency components via mixing

and wave propagation as they meet a south-eastward recirculation (figure not shown). This

is also implied from the fact that the tracer concentration on σ41.50 is about 30% in this

region after 150 yr of release, compared to more than 50% at the bottom-most level (see

Fig. 4 for comparison). Nonetheless, the similarity in the frequency of the along-isopycnal

and along-isobar θ−S anomalies, and hσ anomalies (Fig. 19c), suggests that they are linked

at the source region as set by a common mechanism (section 4). This also implies that the

isopycnal heaving in the interior is a signature of baroclinic wave propagation initiated by

the density perturbation in the Weddell Sea.

24

To establish the link between θ − S anomalies in the interior and the Antarctic over-

turning (ψatl), the variables shown in Fig. 18 are correlated against ψatl at various time

lags (Fig. 20). An increase in ψatl leads to a widespread cooling along isobars (Fig. 20b)

and shoaling of isopycnals (Fig. 20e) across ≈ 100◦ of longitude over ≈ 20 yr. On the

other hand, we see a dipole structure of S anomalies along isobars (Fig. 20a) and of θ − S

anomalies along the σ41.50 isopycnal (Fig. 20c, d). Hence, an increase in ψatl is associated

with warmer and higher salinity water along the isopycnal within the Weddell Sea. The

along-isopycnal cooling and freshening further east correspond to periods of anomalously

weak overturning. Similar patterns are captured along the meridional transect as presented

in Fig. 19, however, they are apparent only south of 40◦S. North of this latitude, the θ− S

variability is over time scales that are too long to be statistically resolved by this analysis.

The difference in ψatl correlation patterns with θ−S on isobars and isopycnals suggest that

care should be taken when associating θ − S anomalies in the interior to fluctuations in

AABW formation and outflow.

Finally, it is worth mentioning that the magnitude of the maximum decadal θ − S

changes along isopycnals in the model is of the order of 0.1◦C, 0.01 psu in the Weddell

Sea and 0.01◦C, 0.001 psu in the Atlantic interior near 43◦S. These are comparable to the

magnitude of θ−S changes on decadal time scales found in observations (Coles et al. 1996;

Meredith et al. 2001; Fahrbach et al. 2004) and the study by Stossel and Kim (2001).

6. Summary and Conclusions

The natural variability of AABW has been analysed in a coupled climate model. Ex-

amination of passive tracer concentrations suggests the model AABW is predominantly

25

sourced in the Weddell Sea, with weak contribution from the Ross Sea and insignificant

bottom water formation in the Adelie Land region, the latter in contrast to observations.

This deficiency is common in coarse resolution GCMs, likely due to inadequate representa-

tion of dense water overflow, convective processes, and surface boundary conditions in that

sector of the Southern Ocean. In contrast, the model successfully reproduces key features

of bottom water pathways in the Atlantic sector. The focus of the present study is therefore

on the Atlantic sector AABW, sourced by Weddell Sea Deep Water (WSDW) and Weddell

Sea Bottom Water (WSBW).

The Atlantic sector Antarctic overturning (ψatl) was shown to effectively approximate

Weddell Sea deep and bottom water transport rates in the model. Thus, we examined

variability in ψatl to investigate the variability of WSDW/WSBW formation and outflow.

The overturning variability is tightly linked to surface density anomalies in the Weddell Sea,

with strong and weak phases of ψatl characterised by composite patterns that are mirror

image of each other. During phases of strong overturning anomalies, sea surface density

first increases approximately 2 yr prior to the increase in overturning, and then continues

to increase with depth over the next decade. The increase in surface density is accompanied

by positive θ − S anomalies, indicating that salinity variations control the fluctuations in

the overturning, not temperature. The spectral peaks of ψatl were shown to closely match

those of SSS over various time scales, while coinciding with SST at only 20 and 30-yr

periods. This implies that the Weddell Sea surface salinity, not temperature, should be

used in paleo-reconstructions of AABW variability, akin to SST for the North Atlantic

thermohaline circulation as shown by Latif et al. (2004). It should be noted, however, that

the amount of salt required to form AABW must also vary over much longer time scales,

26

as a function of the mean global ocean salinity, which is in turn related to the net volume

(and sea level) of the ocean. Thus, this AABW−SSS link has more direct implications for

the use of surface salinity in reconstructing AABW formation rates for the past 7000 years,

when the global sea level has been near its present-day value (Fleming et al. 1998).

The changes in convection triggered by surface salinity anomalies transmit the salinity

signal to depth. However, temperature anomalies below the mixed layer are of opposite sign

than those at the surface. This is a result of convective processes injecting cold and fresh

water downward. It is noted that while surface density is ultimately controlled by salinity,

the dominance of salinity on density variations at depth is moderated, while temperature

variations become more apparent. Nonetheless, the surface salinity anomalies prevail at

depth, but with a reduced amplitude due to damping by convective mixing. A similar

vertical structure of θ − S variability on interdecadal timescales was found in an idealised

ocean-only model forced by surface mixed boundary conditions (Arzel et al. 2006). Arzel

et al. (2006) explain that the anomalous upward injection of warm and saline waters

enhances the positively correlated surface θ− S anomalies, and thus the growth of density

anomalies (i.e., a positive feedback mechanism). However, no explanation is offered for

what drives the periodic oscillation from positive to negative anomalies in their model. The

presence of a sea-ice component in our model provides a source of freshwater flux anomalies

coupled to sub-surface heating of WDW inflow. The increased inflow of WDW following

vigorous overturning, provides surface heat anomalies inducing sea-ice melting and thus

the anomalous freshening of surface waters. The overturning is then reduced, leading to

a suppressed inflow of WDW, inhibiting sea-ice melting. This internal negative-feedback

loop spikes overturning oscillations on interdecadal time scales (see Figs. 13, 14).

27

Direct atmospheric forcing was found to play an important role in generating positively-

correlated θ−S anomalies in the Weddell Sea, thus initiating the internal negative-feedback

mechanism described above. The dominance of the solar heat flux and ice-ocean equivalent

salt flux was revealed by a heat and salt budget analysis of the region. Employing composite

analyses, we deduced that enhanced westerlies drive enhanced sea-ice drift, resulting in a

reduction of sea ice available for summer melting. This results in a positive ice-ocean salt

flux anomaly and a larger ice-free area for increased absorption of solar radiation by the

ocean. Approaching winter, the larger than normal ice-free area exposed to atmospheric

cooling leads to higher than average ice re-growth and ensuing brine rejection. This mech-

anism results in higher temperature and salinity in the region, leading to years of enhanced

overturning. The processes described above were summarised by the schematic shown in

Fig. 14, which can be naturally induced by known climate modes. Interestingly, a signature

of the Southern Annular Mode (SAM) was revealed by the composite patterns of the zonal

winds during years of Weddell Sea salinity anomalies. Here, positive SAM events correspond

to increased Weddell Sea surface salinity and generally an increased sea-ice extent (see also

Hall and Visbeck 2002; Sen Gupta and England 2006). A modest but significant correlation

between Weddell Sea surface salinity and the characteristic time series of the SAM in zonal

winds implies that SAM events contribute to the model’s bottom water variability.

It is of particular importance to interpret θ − S changes in the interior in association

with overturning fluctuations. To approach this issue, we first presented CEOF analyses

of θ − S anomalies along an isopycnal surface that intercepts the upper layer of WSDW.

The propagation of anomalies was then investigated further along cross-sections within the

South Atlantic. The CEOF analyses reveal a θ−S dipole pattern emerging in the Weddell

28

Sea on various time scales. The anomalies propagate eastward and then northward into

the Atlantic. Fluctuations on interannual to interdecadal time scales are noted to be more

prominent in deeper layers, as the deeper bottom waters are more rapidly ventilated. Along

isobars, salinity variability was shown to be dominated by along-isopycnal propagation of

S anomalies. On the other hand, θ variability is dominated by signatures of isopycnal

displacements exhibiting propagation of density anomalies with the speed of baroclinic

waves (see also Stossel and Kim 2001). A lagged-correlation analysis between θ−S against

the overturning reveals a basin-scale uniform θ anomaly pattern, in contrast to the dipole

pattern of salinity. An increase in the overturning is associated with a widespread cooling

and shoaling of isopycnals, higher salinity in the western-central Weddell Sea, and lower

salinity in the eastern Weddell Sea.

Finally, we note that although the mechanisms described here could potentially be the

dominant mechanisms driving AABW variability in the real system, some of the results

are likely to be sensitive to model parameters such as the resolution and mixing parame-

terisation employed. For instance, the propagation of anomalies would likely be faster in

higher-resolution models, as ocean currents and topographic features are better resolved.

Inclusion of a parameterisation of downslope flows could also potentially affect the time

scales of variability, and further enhance the role of surface waters on AABW variability.

Ideally, we would have used a fully-coupled global climate model with sufficiently high

resolution to explicitly resolve these processes and coastal polynyas, which is not compu-

tationally feasible at present. Nonetheless, our study highlights, in the broadest sense, the

close link between sea ice, surface salinity, and bottom water variability via atmospheric

forcing and internal feedback mechanisms. Our study on AABW variability is an advance

29

on previous studies given the fully coupled nature of the model, combined with its very

long integration time.

The link between Antarctic sea-ice variability and Southern Hemisphere climate modes

has received increased attention in recent years (e.g., Fichefet et al. 2003; Liu et al. 2004;

Lefebvre et al. 2004; Sen Gupta and England 2006). Our study demonstrates the interplay

between air-sea and ice-sea fluxes, sea surface temperature-salinity, and internal oceanic

advection in setting the magnitude and time-scales of AABW variability. It will be impor-

tant to continue this effort in understanding how regional to global climate modes control

Antarctic sea ice and bottom water variability, particularly as atmospheric greenhouse gas

concentrations continue to rise.

Acknowledgements

The authors thank Mark Collier for preparing the model data output, and Barrie Hunt

and Tony Hirst for access to the 10,000 year climate model simulations. Siobhan O’Farrell

is gratefully acknowledged for providing the passive tracer data used in Figs. 1, 4. The

authors also thank Steve Rintoul, Neil Holbrook, and Siobhan O’Farrell for their helpful

comments. Comments and suggestions by two anonymous reviewers helped improve the

manuscript. This research was supported by the Australian Research Council and the

Australian Antarctic Science Program.

30

References

Arzel O., T. Huck and A. C. de Verdiere, 2006: The different nature of the interdecadal

variability of the thermohaline circulation under mixed and flux boundary conditions.

J. Phys. Oceanogr., 36, 1703−1718.

Baines, P. G., and S. Condie, 1998: Observations and modelling of Antarctic downslope

flows: A review. In: Ocean, Ice, and Atmosphere: Interactions at the Antarctic

Continental Margin, Antarctic Research Series, Vol. 75, edited by S. S. Jacobs and

R. F. Weiss, p. 29−49, AGU, Washington DC.

Brix, H., and R. Gerdes, 2003: North Atlantic Deep Water and Antarctic Bottom Water:

their interaction and influence on the variability of the global ocean circulation. J.

Geophys. Res., 108, doi:10.1029/2002JC001335.

Bryan, K., 1969: A numerical method for the study of the circulation of the World Ocean.

J. Comput. Phys., 3, 347−376.

Carmack, E. C., 1977: Water characteristics of the Southern Ocean south of the Polar

Front. In: A Voyage of Discovery, edited by M. Angel, Deep-Sea Res., 24(Suppl.).

Coles, V. J., M. S. McCartney, D. B. Olson, and W. M. Smethie Jr., 1996: Changes in the

Antarctic Bottom Water properties in the western South Atlantic in the late 1980s.

J. Geophys. Res., 101, 8957−8970.

Cox, M. D., 1984: A primitive equation, three-dimensional model of the ocean. GFDL

Ocean Group Tech. Rep. No. 1. GFDL. Princeton University, Princeton, NJ, 141 p.

31

Cox, M. D., 1987: Isopycnal diffusion in a z-coordinate ocean model. Ocean Modelling

(unpublished manuscript), 74, 1−15.

Doney, S. C., and M. W. Hecht, 2002: Antarctic Bottom Water formation and deep-water

chlorofluorocarbon distributions in a global ocean climate model. J. Phys. Oceanogr.,

32, 1642−1666.

Drijfhout, S., C. Heinze, M. Latif, and E. Maier-Reimer, 1996: Mean circulation and

internal variability in an ocean primitive equation model. J. Phys. Oceanogr., 26,

559−580.

England, M. H., 1993: Representing the global-scale water masses in ocean general circu-

lation models. J. Phys. Oceanogr., 22, 1523−1552.

England, M. H., and G. Holloway, 1998: Simulations of CFC content and water-mass age

in the deep North Atlantic. J. Geophys. Res., 103, 15,885−15,902.

England, M. H., O. Arzel, A. Santoso, A. Sen Gupta, W. P. Sijp, J. Trevena, 2007: On

the diagnosis of Antarctic Bottom Water overturning rates in OGCMs. J. Phys.

Oceanogr., in preparation.

Fahrbach, E., M. Hoppema, G. Rohardt, M. Schroder, and A. Wisotzki, 2004: Decadal-

scale variations of water mass properties in the deep Weddell Sea. Ocean Dynamics,

54, 77−91.

Fichefet, T., B. Tartinville, and H. Goosse, 2003: Antarctic sea ice variability during

1958−1999: A simulation with a global ice-ocean model. J. Geophys. Res., 108,

doi:10.1029/2001JC001148.

32

Flato, G. M., and W. D. Hibler, 1990: On a simple sea-ice dynamics model for climate

studies. Ann. Glaciol., 14, 72−77.

Fleming, K., P. Johnston, D. Zwartz, Y. Yokoyama, K. Lambeck, and J. Chappell, 1998:

Refining the eustatic sea-level curve since the Last Glacial Maximum using far- and

intermediate-field sites. Earth and Planetary Science Letters, 163, 327−342.

Foldvik, A., T. Gammelsrød, S. Østerhus, E. Fahrbach, G. Rohardt, M. Schroder, K.

W. Nicholls, L. Padman, and R. A. Woodgate, 2004: Ice shelf water overflow and

bottom water formation in the southern Weddell Sea. J. Geophys. Res., 109,

doi:10.1029/2003JC002008.

Foster, T. D., and E. C. Carmack, 1976: Frontal zone mixing and Antarctic Bottom Water

formation in the southern Weddell Sea. Deep-Sea Res., 23, 301−317.

Gent, P. R., and J. C. McWilliams, 1990: Isopycnal mixing in ocean circulation models.

J. Phys. Oceanogr., 20, 150−155.

Gent, P. R., J. Willebrand, T. J. McDougall, and J. C. McWilliams, 1995: Parameterizing

eddy-induced tracer transports in ocean circulation models. J. Phys. Oceanogr., 25,

463−474.

Gill, A. E., 1973: Circulation and bottom water formation in the Weddell Sea. Deep-Sea

Res., 20, 111−140.

Goosse, H., J.-M. Campin, B. Tartinville, 2001: The sources of Antarctic bottom water in

a global ice-ocean model. Ocean Modelling, 3, 51−65.

33

Gordon, H. B., and S. P. O’Farrell, 1997: Transient climate change in the CSIRO coupled

model with dynamic sea ice. Mon. Wea. Rev., 125, 875−907.

Haine, T. W. N., A. J. Watson, M. I. Liddicoat, R. R. Dickson, 1998: The flow of Antarctic

bottom water to the southwest Indian Ocean estimated using CFCs. J. Geophys. Res.,

103, 27637−27654.

Hall, A., and M. Visbeck, 2002: Synchronous variability in the Southern Hemisphere

atmosphere, sea ice, and ocean resulting from the Annular Mode. J. Climate, 15,

3043−3057.

Harms, S., E. Fahrbach, and V. H. Strass, 2001: Sea ice transports in the Weddell Sea. J.

Geophys. Res., 106, 9057−9073.

Hirst, A. C., and T. J. McDougall, 1996: Deep-water properties and surface buoyancy flux

as simulated by a Z -coordinate model including eddy-induced advection. J. Phys.

Oceanogr., 26, 1320−1343.

Hirst, A. C., S. P. O’Farrell, and H. B. Gordon, 2000: Comparison of a coupled ocean-

atmosphere model with and without oceanic eddy-induced advection. Part I: ocean

spinup and control integrations. J. Climate, 13, 139−163.

Hogg, N. G., and W. Zenk, 1997: Long-period changes in the bottom water flowing through

Vema Channel. J. Geophys. Res., 102, 15639−15646.

Hogg, N. G., G. Siedler, and W. Zenk, 1999: Circulation and variability at the southern

boundary of the Brazil Basin. J. Phys. Oceanogr., 29, 145−157.

34

Hunt, B. G., 2004: The stationarity of global mean climate. Int. J. Climatol., 24,

795−806.

Jacobs, S. S., R. G. Fairbanks, and Y. Horibe, 1985: Origin and evolution of water masses

near the Antarctic continental margin: evidence from H218O/H2

16O ratios in seawater.

Antarctic Research Series, 43, 59−85.

Jacobs, S. S., 2004: Bottom water production and its links with the thermohaline circula-

tion. Antarctic Science, 4, 427−437.

Johnson, G. C., and S. C. Doney, 2006: Recent western South Atlantic bottom water

warming. Geophys. Res. Lett., 33, doi:10.1029/2006GL026769.

Kowalczyk, E. A., J. R. Garratt, and P. B. Krummel, 1994: Implementation of a soil

canopy scheme into the CSIRO GCM − Regional aspects of the model response.

CSIRO Division of Atmospheric Research, Tech. Paper 32, 59 p. [Available from

CSIRO Atmospheric Research, PMB1, Aspendale, Vic. 3195, Australia.]

Larque, L., K. Maamaatuaiahutapu, V. Garcon, 1997: On the intermediate and deep

water flows in the South Atlantic Ocean. J. Geophys. Res., textbf102, 12425−12440.

Latif, M., E. Roeckner, M. Botzet, M. Esch, H. Haak, S. Hagemann, J. Jungclaus, S.

Legutke, S. Marsland, U. Mikolajewicz, and J. Mitchell, 2004: Reconstructing, moni-

toring, and predicting multidecadal-scale changes in the North Atlantic thermohaline

circulation with sea surface temperature. J. Climate, 17, 1605−1614.

Lefebvre, W., H. Goosse, and T. Fichefet, 2004: Influence of the Southern Annular Mode

on the sea ice-ocean system. J. Geophys. Res., 109, doi:10.1029/2004JC002403.

35

Liu, J., J. A., Curry, D. G., Martinson, 2004: Interpretation of recent Antarctic sea ice

variability. Geophys. Res. Lett., 31, doi:10.1029/2003GL018732.

Mantyla, A. W., and J. L. Reid, 1983: Abyssal characteristics of the World Ocean Waters.

Deep-Sea Res., 30, 805−833.

Mantyla, A. W., and J. L. Reid, 1995: On the origins of deep and bottom waters of the

Indian Ocean. J. Geophys. Res., 100, 2417−2439.

Meredith, M. P., R. A. Locarnini, K. A. Van Scoy, A. J. Watson, K. J. Heywood, and B.

A. King, 2000: On the sources of Weddell Gyre Antarctic Bottom Water. J. Geophys.

Res., 105, 1093−1104.

Meredith, M. P., A. C. Naveira Garabato, D. P. Stevens, K. J. Heywood, and R. J. Sanders,

2001: Deep and bottom waters in the Eastern Scotia Sea: Rapid changes in properties

and circulation. J. Phys. Oceanogr., 31, 2157−2168.

Naveira Garabato, A. C., K. J. Heywood, and D. P. Stevens, 2002: Modification and

pathways of Southern Ocean deep waters in the Scotia Sea. Deep-Sea Research I, 49,

681−705.

North, G. R., T. L. Bell, R. F. Cahalan, and F. J. Moeng, 1982: Sampling errors in the

estimation of Empirical Orthogonal Functions. Mon. Wea. Rev., 110, 699−706.

O’Farrell, S.P., 1998: Sensitivity study of a dynamical sea ice model: The effect of the

external stresses and land boundary conditions on ice thickness distribution. J. Geo-

phys. Res., 103, 15751−15782.

36

O’Farrell, S. P., 2002: Use of passive tracers as a diagnostic tool in coupled model simu-

lations - Northern Hemisphere. J. Phys. Oceanogr., 32, 831−850.

Orsi, A. H., G. C. Johnson, and J. L. Bullister, 1999: Circulation, mixing, and production

of Antarctic Bottom Water. Prog. in Oceanogr., 43, 55−109.

Qiu B., W. Miao, and P. Muller, 1997: Propagation and decay of forced and free baroclinic

Rossby waves in off-equatorial oceans. J. Phys. Oceanogr., 27, 2405−2417.

Redi, M. H., 1982: Oceanic isopycnal mixing by coordinate rotation. J. Phys. Oceanogr.,

12, 1154−1158.

Reid, J. L., 1989: On the total geostrophic circulation of the South Atlantic Ocean: flow

patterns, tracers, and transports. Prog. Oceanogr., 23, 149−244.

Rintoul, S. R., 1998: On the origin and influence of Adelie Land Bottom Water. In: Ocean,

Ice, and Atmosphere: Interactions at the Antarctic Continental Margin, Antarctic

Research Series, Vol. 75, edited by S. S. Jacobs and R. F. Weiss, p. 151−171, AGU,

Washington DC.

Rintoul, S. R., and M. H. England, 2002: Ekman transport dominates local air-sea fluxes in

driving variability of Subantarctic Mode Water. J. Phys. Oceanogr., 32, 1308−1320.

Robertson, R., M. Visbeck, A. L. Gordon, and E. Fahrbach, 2002: Long-term temperature

trends in the deep waters of the Weddell Sea. Deep-Sea Res. II, 49, 4791−4806.

Santoso, A., and M. H. England, 2004: Antarctic Intermediate Water circulation and

variability in a coupled climate model. J. Phys. Oceanogr., 34, 2160−2179.

37

Santoso, A., M. H. England, and A. C. Hirst, 2006: Circumpolar Deep Water circulation

and variability in a coupled climate model. J. Phys. Oceanogr., 36, 1523−1552.

Sen Gupta, A. and M. H. England, 2006: Coupled ocean-atmosphere-ice response to

variations in the Southern Annular Mode. J. Climate, 19, 4457−4486.

Semtner, A. J., 1976: A model of the thermodynamics growth of sea ice in numerical

investigations of climate. J. Phys. Oceanogr., 6, 379−389.

Stossel, A., and S.-J. Kim, 1998: An interannual Antarctic sea-ice-ocean mode. Geophys.

Res. Lett., 25, 1007−1010.

Stossel, A., S.-J. Kim, S. S. Drijfhout, 1998: The impact of Southern Ocean sea ice in a

global ocean model. J. Phys. Oceanogr., 28, 1999−2018.

Stossel, A., S. J. Kim, 2001: Decadal deep-water variability in the subtropical Atlantic

and convection in the Weddell Sea. J. Geophys. Res., 106, 22425−22440.

Stossel, A., K. Yang, and S.-J. Kim, 2002: On the role of sea ice and convection in a global

ocean model. J. Phys. Oceanogr., 32, 1194−1208.

Toggweiler, J. R., and B. Samuels, 1995: Effect of sea ice on the salinity of Antarctic

Bottom Water. J. Phys. Oceanogr., 25, 1980−1997.

Uotila, J., T. Vihma, and J. Launiainen, 2000: Response of the Weddell Sea pack ice to

wind forcing. J. Geophys. Res., 105, 1135−1151.

Whitworth III, T., and W. D. Nowlin, 1987: Water masses and currents of the Southern

Ocean at the Greenwich meridian, J. Geophys. Res., 92, 6462−6476.

38

Whitworth III, T., A. H. Orsi, S.-J. Kim, W. D. Nowlin Jr., and R. A. Locarnini, 1998:

Water masses and mixing near the Antarctic slope front. In: Ocean, Ice, and Atmo-

sphere: Interactions at the Antarctic Continental Margin, Antarctic Research Series,

Vol. 75, edited by S. S. Jacobs and R. F. Weiss, p. 1−27, AGU, Washington DC.

Zwally, H. J., J. C. Comiso, C. L. Parkinson, D. J. Cavalieri, and P. Gloersen, 2002: Vari-

ability of Antarctic sea ice 1979−1998. J. Geophys. Res., 107, doi:10.1029/2000JC000733.

39

Table Captions

Table 1: The mean and range of transports within the Weddell Sea. The term ψatl is

the magnitude of the Antarctic overturning cell shown in Fig. 3b. The transports

are calculated within the proximity of the Weddell Sea (between 3◦E and 65◦W).

‘Sinking at 1500 m’ refers to the sinking of bottom water defined as the integral of

downward velocities south of 68◦S at 1500-m depth where ψatl is most rapid. ‘Outflow’

refers to the integral of meridional velocities at 68◦S underneath the σ41.50 surface and

below 1500-m depth (see Eq. 1; refer to Fig. 2), and ‘inflow’ refers to the integral

of meridional velocities from the surface to the depth of the σ41.50 surface. The

maximum correlation coefficients against the Atlantic sector Antarctic overturning

(ψatl) are shown with the indicated time lags (years) in brackets. A positive time lag

indicates ψatl leading the specified variable.

Table 2: Standard deviation of the annual-mean surface heat budget terms over 1000

years spatially averaged over the region indicated in Fig. 5.

Table 3: Standard deviation of surface salinity budget terms. The terms are calculated

as for those in Table 2. For consistency of units, evaporation and precipitation have

been converted to equivalent salt fluxes (psu s−1). Note that the salt flux into the

ocean referred to here is equivalent to a negative freshwater flux in the coupled model.

40

Figure Captions

Figure 1: Passive tracer concentration at the model’s bottom most ocean grid boxes

at 50 yr (top panel) and 150 yr (bottom panel) after release at the surface. The

corresponding mean current velocities at the bottom-most level are shown by the

velocity vectors. The bottom-most ocean grid boxes can correspond to the top of

ridges, but more generally track the abyssal oceans. The model bottom topography

is presented by thick contours marking the 4000-m isobath and thin contours marking

the 3000-m isobath.

Figure 2: Meridional velocity along a circumpolar transect at 68.5◦S. The mean position

of the σ41.50 (σ45.95) isopycnal is marked by solid (dashed) contours (see text for

definition of the σ41.50 and σ45.95 surfaces).

Figure 3: (a) Global meridional overturning circulation (MOC) averaged over 1000 model

years. (b) Atlantic sector MOC. Solid (dashed) contours indicate positive (negative)

overturning in Sv (1 Sv ≡ 106 m3 s−1). The Antarctic overturning cell is highlighted

using bold dashed contours. The mean position of the σ41.50 and σ45.95 isopycnals

are shown in (b) in bold contours. (c) Time series of the maximum magnitude of

the Antarctic overturning cell for the global mean (black) and the Atlantic sector

(gray). The magnitude of the overturning rate and its variability will be analysed in

this study. The horizontal dashed lines indicate one standard deviation above and

below the long-term mean. (d) Time series of the maximum Atlantic sector Antarctic

overturning calculated on the ρ3 vertical level (black) and the z -level counterpart

(gray). The time snap-shot in (d) coincides with the period shown in Fig. 6. Note

41

that in (a), (b) the MOC is derived from advection without inclusion of the GM

terms.

Figure 4: (a) Tracer concentration at 150 yr after release, (b) potential temperature, and

(c) salinity, for the Atlantic sector on σ41.50 versus that on σ45.95 (dots) and for the

bottom-most model level versus σ45.95 concentration (crosses). Weddell Sea Bottom

Water can be taken to lie on σ45.95 whereas WSDW lies on σ41.50. See text for further

details.

Figure 5: Lagged correlations between sea surface density (SSD) and the Atlantic sector

Antarctic overturning (ψatl) in which surface density leads ψatl by the indicated lag

in years. The correlation maps focus on the Weddell Sea region. The correlations are

based on 200 years of model data. Correlations above ≈ 0.2 are significant at the 95%

confidence level.

Figure 6: Time series of the Atlantic sector Antarctic overturning (ψatl) versus (a) SSS

and (b) SST, spatially averaged over the Weddell Sea as indicated by the box in Fig.

5 (denoted as SSSwed and SSTwed hereafter).

Figure 7: Power spectral density of (a) ψatl, (b) SSSwed, and SSTwed (see Fig. 6 cap-

tion for definition of the variables). The dashed curve indicates the fitted red-noise

spectrum at 90% confidence level.

Figure 8: (top) Lagged correlation between the spatially-averaged (a) density, (b) salinity,

and (c) temperature in the Weddell Sea versus ψatl. (bottom) Lagged correlation

between the surface and the regional depth profile of (d) density, (e) salinity, and (f)

temperature. Only correlation values exceeding ±0.2 (above the 95% confidence level)

42

are contoured. Positive correlations are shown in solid contours and gray shading.

Negative correlations are shown in dashed contours. Negative (positive) time lags

indicate the first mentioned variables lead (lag) the latter. The tracer variables are

spatially-averaged over the region indicated in Fig. 5.

Figure 9: θ−S anomalies at various fixed depths in αθ′

−βS′

space at the indicated time

lags of anomalously high (dark dots) and low (light dots) ψatl. The time lags are deter-

mined by the maximum lagged correlations of the density at the specified depth versus

ψatl. The composite averages are shown by the thick dark and light lines. The abbre-

viations WF, WS, CS, CF indicate the signs of the θ−S anomalies, corresponding to

warming-freshening, warming-salination, cooling-salination, and cooling-freshening,

respectively. The line Rρ = 1 at 45◦ angle indicates perfect density compensating

θ − S variations. Details of the analysis can be found in the Appendix of Santoso et

al. (2006). The line at Rρ = −1 indicates maximum density instability.

Figure 10: Composite maps of (left column) annual mean, (middle column) summer,

and (right column) winter mean anomalies of (a) ice concentration, (b) ice-ocean salt

flux, and (c) short-wave radiative flux, calculated based on the high salinity years (i.e.,

when SSSwed exceeds one standard deviation unit above the mean). In (a, b), positive

sign indicates an increase in ice concentration and ice-ocean salt fluxes respectively.

In (c), positive sign indicates an increased solar radiation (i.e., increased heating of

the ocean). The sea-ice salt flux into the ocean referred to here can be equivalently

interpreted as a negative freshwater flux.

Figure 11: Composite maps of annual-mean (a) SSS, (b) SST, (c) zonal wind stress, (d)

43

meridional wind stress, and (e) net air/ice-ocean salt flux based on the (left) high

Weddell Sea salinity years and (right) low Weddell Sea salinity years. The region in

the Weddell Sea used to define high and low salinity years is indicated.

Figure 12: EOF analysis of the annual-mean zonal wind stress (τ x) showing (a) the

spatial map of the first EOF mode and (b) the principal component of the first mode

(PC1) accounting for 26.4% of the total τ x variance. (c) The power spectrum of PC1

with the fitted white-noise background spectrum at 95% confidence level.

Figure 13: Lagged correlation of (a) ψatl versus ice-ocean salt flux, (b) ice-ocean salt

flux versus SSSwed, and (c) SSSwed versus ψatl, using the raw data (thin curve), data

filtered with band-pass period of 10−50 yr (gray curve), and with band-pass period

of 10−33 yr (thick curve). The dashed horizontal lines are the corresponding 95%

significance level coefficients for the band-pass filtered analysis. The dotted line is the

95% significance level for the raw time-series analysis. Negative (positive) time lags

indicate the first mentioned variable leads (lags) the latter variable.

Figure 14: Life cycle of the Weddell Sea overturning anomalies linked to surface salinity

and temperature as described in the text (section 4). (a) Anomalously strong west-

erlies and katabatic winds drive enhanced sea-ice drift, resulting in a reduction of sea

ice available for summer melting, thus creating a positive ice-ocean salt flux anomaly

and increased absorption of solar radiation into the ocean. (b) Approaching winter,

the larger than normal ice-free area exposed to atmospheric cooling leads to higher

than average brine rejection, driving stronger overturning. Conversely, at other times,

weaker wind forcing leads to low overturning as illustrated in (c), (d). In this way,

44

surface variability initiates an internal negative-feedback oscillation. A half oscillation

period is illustrated by the strengthening of bottom water outflow (b) approximately 5

yr after the generation of surface anomalies, leading to stronger inflow of WDW (also

seen in b), which then leads to enhanced heat injection into the mixed layer about 1

yr later (c). This results in enhanced sea-ice melting and thus weaker bottom water

outflow and WDW inflow (d), which makes up the second half of the oscillation. This

internal negative-feedback loop generates an overturning oscillation on interdecadal

time scales (see also Fig. 13).

Figure 15: Standard deviation of θ− S on the σ41.50 isopycnal surface. A log10 scale has

been applied to the colour scheme to enhance signal visibility in the interior.

Figure 16: (a) Spatial maps of the three leading complex EOF modes of salinity shown

at 90◦ phase intervals along σ41.50. The first (CEOF1), second (CEOF2), and third

(CEOF3) modes account for 42.4%, 21.9%, and 12.2% of salinity variance, respec-

tively. An equivalent CEOF analysis of θ shows similar modal structure and so is

not shown here. (b) Spatial phase angle CEOF1−3 indicating the direction of phase

propagation of each mode from 0◦ to 360◦. Apparent phase discontinuities occur be-

cause the phase is defined only between 0◦ and 360◦. The horizontal and vertical

lines shown in the CEOF1 map (top row, middle column) mark the position of the

transects for the Hovmoller diagrams shown in Figs. 18, 19. The box in the middle

and bottom rows (middle column) indicate the Weddell Sea region.

Figure 17: (left column) Principal component time series of CEOF1−3 shown for σ41.50

(black) and σ45.95 (gray). (right column) The corresponding power spectral density

45

(PSD) of the principal component time series with an estimated autoregressive-1

background spectrum at 90% confidence level (dashed curves). A log10 scale is used

in frequency to give more weight to the higher frequency signals and the power spectra

are multiplied by frequency to preserve variance.

Figure 18: Hovmoller diagram of (a, b) θ − S anomalies along isopycnal, (c) isopycnal

depth anomalies, and (d, e) θ − S anomalies along isobars on σ41.50 along a zonal

transect at 62◦S from 60◦W to 87◦E (see bottom right inset for the transect location).

The along-isobar θ−S anomalies are calculated as deviations from the 1000-yr mean

along the mean isopycnal depth of σ41.50.

Figure 19: As in Fig. 18, but for the meridional transect shown in the bottom right

inset.

Figure 20: Lagged correlation of Atlantic overturning (ψatl) vs (a, b) S − θ at the mean

depth of the σ41.50 isopycnal surface, (c, d) S − θ on σ41.50, and (e) the depth of

the σ41.50 surface along the zonal transect shown in Fig. 18. Positive (negative)

correlations are shown by solid (dashed) contours. Positive time lags indicate ψatl

leading the respective hydrographic variables.

46

Table 1: The mean and range of transports within the Weddell Sea. The term ψatl is the

magnitude of the Antarctic overturning cell shown in Fig. 3b. The transports are calculated

within the proximity of the Weddell Sea (between 3◦E and 65◦W). ‘Sinking at 1500 m’ refers to

the sinking of bottom water defined as the integral of downward velocities south of 68◦S at 1500-m

depth where ψatl is most rapid. ‘Outflow’ refers to the integral of meridional velocities at 68◦S

underneath the σ41.50 surface and below 1500-m depth (see Eq. 1; refer to Fig. 2), and ‘inflow’

refers to the integral of meridional velocities from the surface to the depth of the σ41.50 surface.

The maximum correlation coefficients against the Atlantic sector Antarctic overturning (ψatl) are

shown with the indicated time lags (years) in brackets. A positive time lag indicates ψatl leading

the specified variable.

Mean Range Correlation

(Sv) (Sv) against ψatl

Atlantic overturning (ψatl) 8.6 4.5 1

Sinking at 1500 m 8.0 3.6 0.99 (0 yr)

Outflow 8.0 4.0 0.97 (0 yr)

Inflow −8.0 3.8 −0.90 (1 yr)

47

Table 2: Standard deviation of the annual-mean surface heat budget terms over 1000 years

spatially averaged over the region indicated in Fig. 5.

Budget term std dev

(×10−8◦C s−1)

∂θ/∂t 0.37

Zonal advection (u∂θ/∂x) 0.10

Meridional advection (v∂θ/∂y) 0.18

Vertical advection (w∂θ/∂z) 5.59×10−5

Net surface heat flux (Qnet) 1.20

Air to ocean solar radiation (Qsolar) 2.88

Ocean to air long wave radiation (Qlw) 1.25

Ocean to air sensible heat flux (Qsh) 1.02

Ocean to air evaporative heat flux (Qevp) 0.64

Ocean to ice sensible heat flux (Qoi) 0.69

48

Table 3. Standard deviation of surface salinity budget terms. The terms are calculated as for

those in Table 2. For consistency of units, evaporation and precipitation have been converted to

equivalent salt fluxes (psu s−1). Note that the salt flux into the ocean referred to here is equivalent

to a negative freshwater flux in the coupled model.

Budget term std dev

(×10−9 psu s−1)

∂S/∂t 0.93

Zonal advection (u∂S/∂x) 0.24

Meridional advection (v∂S/∂y) 0.47

Vertical advection (w∂S/∂z) 3.56×10−4

Net air/ice-ocean salt flux (Hnet) 5.53

Evaporative salt flux (E) 0.36

Precipitative salt flux (P ) 1.47

Ice to ocean salt flux (Hice) 7.01

49

Latit

ude

80S

60S

40S

20S

0

Longitude

Latit

ude

150E 160W 110W 60W 10W 40E 90E

80S

60S

40S

20S

0

(%)0 10 20 30 40 50 60 70 80 90 100

t=50 yr

t=150 yr

Figure 1: Passive tracer concentration at the model’s bottom most ocean grid boxes at 50yr (top panel) and 150 yr (bottom panel) after release at the surface. The correspondingmean current velocities at the bottom-most level are shown by the velocity vectors. Thebottom-most ocean grid boxes can correspond to the top of ridges, but more generally trackthe abyssal oceans. The model bottom topography is presented by thick contours markingthe 4000-m isobath and thin contours marking the 3000-m isobath.

50

Meridional velocity at 68.5°S

Longitude

Dep

th (

m)

150E 110W 10W 90E4500

3500

2500

1500

500

(cm s−1)

−1 −0.5 0 0.5 1

Figure 2: Meridional velocity along a circumpolar transect at 68.5◦S. The mean position ofthe σ41.50 (σ45.95) isopycnal is marked by solid (dashed) contours (see text for definition ofthe σ41.50 and σ45.95 surfaces).

51

1310413

4−10

−4−1

−7

−1

−4

Dep

th (

m)

(a) Global meridional overturning

80S 60S 40S 20S 0 20N 40N 60N 80N

−4000

−3000

−2000

−1000

0

−1

−3

−7

−3

−1

139

5

11

Latitude

Dep

th (

m)

(b) Atlantic meridional overturning

80S 60S 40S 20S 0 20N 40N 60N 80N

−4000

−3000

−2000

−1000

0

0 100 200 300 400 500 600 700 800 900 10006

7

8

9

10

11

Tra

nspo

rt (

Sv)

(c) Antarctic overturning rate (ψatl

)

GlobalAtlantic

700 720 740 760 780 800 820 840 860 8806

7

8

9

10

11

Time (year)

(Sv)

ψatl

ψatl

|σ

(d) ψatl

vs ψatl

|σ

r = 0.86 (1 yr)3

4

5

6

7

8

(Sv)

σ41.50

σ45.95

ψatl

AABWprod−uction

Figure 3: (a) Global meridional overturning circulation (MOC) averaged over 1000 modelyears. (b) Atlantic sector MOC. Solid (dashed) contours indicate positive (negative) over-turning in Sv (1 Sv ≡ 106 m3 s−1). The Antarctic overturning cell is highlighted using bolddashed contours. The mean position of the σ41.50 and σ45.95 isopycnals are shown in (b)in bold contours. (c) Time series of the maximum magnitude of the Antarctic overturn-ing cell for the global mean (black) and the Atlantic sector (gray). The magnitude of theoverturning rate and its variability will be analysed in this study. The horizontal dashedlines indicate one standard deviation above and below the long-term mean. (d) Time seriesof the maximum Atlantic sector Antarctic overturning calculated on the ρ3 vertical level(black) and the z -level counterpart (gray). The time snap-shot in (d) coincides with theperiod shown in Fig. 6. Note that in (a), (b) the MOC is derived from advection withoutinclusion of the GM terms.

52

0 25 50 75 1000

25

50

75

100

σ45.95

(%)

(%)

σ41.50

bottom

−2 −1 0 1 2−2

−1

0

1

2

σ45.95

(°C)

(° C)

σ41.50

bottom

34.3 34.4 34.5 34.6

34.4

34.6

34.8

σ45.95

(psu)

(psu

)

σ41.50

bottom

(a) Tracer concentration

(b) Potential temperature

(c) Salinity

Figure 4: (a) Tracer concentration at 150 yr after release, (b) potential temperature, and(c) salinity, for the Atlantic sector on σ41.50 versus that on σ45.95 (dots) and for the bottom-most model level versus σ45.95 concentration (crosses). Weddell Sea Bottom Water can betaken to lie on σ45.95 whereas WSDW lies on σ41.50. See text for further details.

53

0 0.2 0.4 >0.6

lag=4 yr

lag=2 yr

lag=0 yr

Figure 5: Lagged correlations between sea surface density (SSD) and the Atlantic sectorAntarctic overturning (ψatl) in which surface density leads ψatl by the indicated lag in years.The correlation maps focus on the Weddell Sea region. The correlations are based on 200years of model data. Correlations above ≈ 0.2 are significant at the 95% confidence level.

54

700 720 740 760 780 800 820 840 860 8806

7

8

9

10

11

(Sv)

700 720 740 760 780 800 820 840 860 88034

34.05

34.1

34.15

34.2

34.25

(psu

)

700 720 740 760 780 800 820 840 860 8806

7

8

9

10

11

Time (year)

(Sv)

700 720 740 760 780 800 820 840 860 880−1.5

−1.32

−1.14

−0.96

−0.78

−0.6

(° C)

(a) ψatl

vs SSSwed

(b) ψatl

vs SSTwed

ψatl

SSSwed

ψatl

SSTwed

r = 0.78

r = 0.52

Figure 6: Time series of the Atlantic sector Antarctic overturning (ψatl) versus (a) SSS and(b) SST, spatially averaged over the Weddell Sea as indicated by the box in Fig. 5 (denotedas SSSwed and SSTwed hereafter).

55

0.001 0.01 0.10

0.25

0.5

PS

D (

Sv2 )

0.001 0.01 0.10

0.5

1

1.5

PS

D (

x10−

3 psu

2 )

0.001 0.01 0.10

0.01

0.02

Frequency (cpy)

PS

D (

° C2 )

89 yr

50 yr

32 yr

21 yr

15 yr

(a) Overturning

(b) SSS

(c) SST

Figure 7: Power spectral density of (a) ψatl, (b) SSSwed, and SSTwed (see Fig. 6 captionfor definition of the variables). The dashed curve indicates the fitted red-noise spectrum at90% confidence level.

56

0.20.4

0.6

0.60.8

Dep

th (

m)

(a) ρ vs overturning

−20 0 20

4000

3000

2000

1000−0.6

−0.4

−0.6

−0.4

−0.2

−0.2(c) θ vs overturning

−20 0 20

0.5

0.4

0.3

0.4

0.70.5

(b) S vs overturning

−20 0 20

0.5

0.4

0.2

0.7

0.5

Dep

th (

m)

(d) ρ vs surface ρ

−20 0 20

4000

3000

2000

1000

0.3

0.5

0.2

Time lag (yr)

(e) S vs SSS

−20 0 20

0.2

0.2

(f) θ vs SST

−0.2

−20 0 20

Figure 8: (top) Lagged correlation between the spatially-averaged (a) density, (b) salinity,and (c) temperature in the Weddell Sea versus ψatl. (bottom) Lagged correlation betweenthe surface and the regional depth profile of (d) density, (e) salinity, and (f) temperature.Only correlation values exceeding ±0.2 (above the 95% confidence level) are contoured.Positive correlations are shown in solid contours and gray shading. Negative correlationsare shown in dashed contours. Negative (positive) time lags indicate the first mentionedvariables lead (lag) the latter. The tracer variables are spatially-averaged over the regionindicated in Fig. 5.

57

0

WS90

WF

180

CF−90

CS

0

WS90

WF

180

CF−90

CS

0

WS90

WF

180

CF−90

CS

0

WS90

WF

180

CF−90

CS

(a) 12.5 m, −3 yr (b) 410 m, −1 yr (c) 2125 m, 0 yr (d) 4375 m, 11 yr

Figure 9: θ − S anomalies at various fixed depths in αθ′

− βS′

space at the indicated timelags of anomalously high (dark dots) and low (light dots) ψatl. The time lags are determinedby the maximum lagged correlations of the density at the specified depth versus ψatl. Thecomposite averages are shown by the thick dark and light lines. The abbreviations WF,WS, CS, CF indicate the signs of the θ−S anomalies, corresponding to warming-freshening,warming-salination, cooling-salination, and cooling-freshening, respectively. The line Rρ =1 at 45◦ angle indicates perfect density compensating θ−S variations. Details of the analysiscan be found in the Appendix of Santoso et al. (2006). The line at Rρ = −1 indicatesmaximum density instability.

58

(x 100%)

−0.05

0

0.05

(psu s−1)

−1

0

1

x 10−8

(W m−2)

−5

0

5

(a)

(b)

(c)

Ice concentration anomaly

Ice−ocean salt flux anomaly

short−wave radiative flux anomaly

Annual Mean Summer Winter

Figure 10: Composite maps of (left column) annual mean, (middle column) summer, and(right column) winter mean anomalies of (a) ice concentration, (b) ice-ocean salt flux, and(c) short-wave radiative flux, calculated based on the high salinity years (i.e., when SSSwed

exceeds one standard deviation unit above the mean). In (a, b), positive sign indicatesan increase in ice concentration and ice-ocean salt fluxes respectively. In (c), positive signindicates an increased solar radiation (i.e., increased heating of the ocean). The sea-ice saltflux into the ocean referred to here can be equivalently interpreted as a negative freshwaterflux.

59

(x 10−9 psu s−1)

−5

0

5

(x 10−9 psu s−1)

−5

0

5

(x 10−3 N m−2)

−2

0

2

(x 10−3 N m−2)

−2

0

2

(x 10−3 N m−2)

−5

0

5

(x 10−3 N m−2)

−5

0

5

(°C)

−0.2

0

0.2

(°C)

−0.1

0

0.1

(psu)

−0.1

0

0.1

(psu)−0.1

0

0.1

High Salinity WS Low Salinity WS

(a) SSS

(b) SST

(c) τx

(d) τy

(e) Hnet

Figure 11: Composite maps of annual-mean (a) SSS, (b) SST, (c) zonal wind stress, (d)meridional wind stress, and (e) net air/ice-ocean salt flux based on the (left) high WeddellSea salinity years and (right) low Weddell Sea salinity years. The region in the WeddellSea used to define high and low salinity years is indicated.

60

(x 10−2 N m−2)−5 0 5

0 250 500 750 1000

−5

0

5

(x10

−2 N

m−

2 )

Time (year)

(b) PC1 (26.4%)

0 0.05 0.1 0.15 0.20

0.5

1

(x10

−3 N

2 m−

4 cpy−

1 )

Frequency (cpy)

(c) Power spectrum

(a) EOF 1

30.3 yr 8.3 yr 95%

Figure 12: EOF analysis of the annual-mean zonal wind stress (τ x) showing (a) the spatialmap of the first EOF mode and (b) the principal component of the first mode (PC1)accounting for 26.4% of the total τ x variance. (c) The power spectrum of PC1 with thefitted white-noise background spectrum at 95% confidence level.

61

Cor

rela

tion

coef

.

Overturning vs ice salt flux

raw10−50 yr10−33 yr

−20 −10 0 10 20

−0.4

0

0.4C

orre

latio

n co

ef.

Ice salt flux vs SSS

raw10−50 yr10−33 yr

−20 −10 0 10 20

−0.4

0

0.4

Time lag (year)

Cor

rela

tion

coef

.

SSS vs overturning

raw10−50 yr10−33 yr

−20 −10 0 10 20−0.8

−0.4

0

0.4

0.8

(a)

(b)

(c)

Figure 13: Lagged correlation of (a) ψatl versus ice-ocean salt flux, (b) ice-ocean salt fluxversus SSSwed, and (c) SSSwed versus ψatl, using the raw data (thin curve), data filteredwith band-pass period of 10−50 yr (grey curve), and with band-pass period of 10−33 yr(thick curve). The dashed horizontal lines are the corresponding 95% significance levelcoefficients for the band-pass filtered analysis. The dotted line is the 95% significance levelfor the raw time-series analysis. Negative (positive) time lags indicate the first mentionedvariable leads (lags) the latter variable.

62

Mixed layer

Summer

(a)

strong Ekman transport

strong katabatic winds

strong westerlies Winter

warm/ saline anomaly

(b)

high overturning

high brine rejection

Summer

(c)

weak westerlies weak katabatic winds

Winter

cold/ fresh anomaly

low brine rejection

Figure 14: Life cycle of the Wetemperature anomalies as descriand katabatic winds drive enhanfor summer melting, thus creaabsorption of solar radiation intice-free area exposed to atmosdriving stronger overturning. Coverturning as illustrated in (cnegative-feedback oscillation. Abottom water outflow (b) approxto stronger inflow of WDW (alsthe mixed layer about 1 yr laterbottom water outflow and WDWThis internal negative-feedback scales (see also Fig. 13).

WDW heat

63

(d)

low overturning

ddell Sea overturning anomalies linked to surface salinity and bed in the text (section 4). (a) Anomalously strong westerlies ced sea-ice drift, resulting in a reduction of sea ice available ting a positive ice-ocean salt flux anomaly and increased o the ocean. (b) Approaching winter, the larger than normal pheric cooling leads to higher than average brine rejection, onversely, at other times, weaker wind forcing leads to low

), (d). In this way, surface variability initiates an internal half oscillation period is illustrated by the strengthening of imately 5 yr after the generation of surface anomalies, leading o seen in b), which then leads to enhanced heat injection into

(c). This results in enhanced sea-ice melting and thus weaker inflow (d), which makes up the second half of the oscillation. loop generates an overturning oscillation on interdecadal time

(a) SD θ (σ41.50

)

Longitude

Latit

ude

150E 110W 10W 90E

80S

60S

40S

20S

0

(x10−2 °C)

1 2 3 4 6 8 11

(b) SD S (σ41.50

)

Longitude

Latit

ude

150E 110W 10W 90E

80S

60S

40S

20S

0

(x10−3 psu)

1 2 3 4 5 6 7 8 9

Figure 15: Standard deviation of θ − S on the σ41.50 isopycnal surface. A log10 scale hasbeen applied to the colour scheme to enhance signal visibility in the interior.

64

CEOF1 (0)

(a)

80S

40S

0

CEOF1 (90)

CEOF1−3 spatial maps

CEOF1 (180)

CEOF2 (0)

Latit

ude

80S

40S

0

CEOF2 (90) CEOF2 (180)

CEOF3 (0)

150E 110W 10W 90E

80S

40S

0

CEOF3 (90)

Longitude150E 110W 10W 90E

−1 0 1

CEOF3 (180)

150E 110W 10W 90E

CEOF1

(b) Spatial phase

CEOF2

Longitude

CEOF3

150E 110W 10W 90E

0 180 360

Figure 16: (a) Spatial maps of the three leading complex EOF modes of salinity shownat 90◦ phase intervals along σ41.50. The first (CEOF1), second (CEOF2), and third(CEOF3) modes account for 42.4%, 21.9%, and 12.2% of salinity variance, respectively.An equivalent CEOF analysis of θ shows similar modal structure and so is not shown here.(b) Spatial phase angle CEOF1−3 indicating the direction of phase propagation of eachmode from 0◦ to 360◦. Apparent phase discontinuities occur because the phase is definedonly between 0◦ and 360◦. The horizontal and vertical lines shown in the CEOF1 map(top row, middle column) mark the position of the transects for the Hovmoller diagramsshown in Figs. 18, 19. The box in the middle and bottom rows (middle column) indicatethe Weddell Sea region.

65

PC1

σ45.95

σ41.50

0 250 500 750 1000−0.02

−0.01

0

0.01

0.02PSD PC1

0.001 0.01 0.10

1

2

3

PC2

S (

psu)

0 250 500 750 1000−0.02

−0.01

0

0.01

0.02PSD PC2

PS

D (

x 10

−5 p

su2 )

0.001 0.01 0.10

0.5

1

1.5

PC3

Time (year)0 250 500 750 1000

−0.02

−0.01

0

0.01

0.02PSD PC3

Frequency (cpy)0.001 0.01 0.10

1

2

77−91 yr 333 yr

333 yr

34.5 yr

Figure 17: (left column) Principal component time series of CEOF1−3 shown for σ41.50

(black) and σ45.95 (grey). (right column) The corresponding power spectral density (PSD)of the principal component time series with an estimated autoregressive-1 background spec-trum at 90% confidence level (dashed curves). A log10 scale is used in frequency to give moreweight to the higher frequency signals and the power spectra are multiplied by frequencyto preserve variance.

66

Tim

e (y

r)

(a) S|σ

60W 10W 40E

200

400

600

800

1000

(psu x10−3)<−5

0

>5(b) θ|σ

60W 10W 40E (°C)<−0.05

0

>0.05(c) hσ

60W 10W 40E (m)<−200

0

>200

Longitude

Tim

e (y

r)

(d) S|z

60W 10W 40E

200

400

600

800

1000

(psu x10−3)<−5

0

>5

Transect location

Longitude

Latit

ude

150E 110W 10W 90E

80S

40S

0

Longitude

(e) θ|z

60W 10W 40E (°C)<−0.05

0

>0.05

Figure 18: Hovmoller diagram of (a, b) θ − S anomalies along isopycnal, (c) isopycnaldepth anomalies, and (d, e) θ − S anomalies along isobars on σ41.50 along a zonal transectat 62◦S from 60◦W to 87◦E (see bottom right inset for the transect location). The along-isobar θ− S anomalies are calculated as deviations from the 1000-yr mean along the meanisopycnal depth of σ41.50.

67

Tim

e (y

r)

(a) S|σ

70S 50S

200

400

600

800

1000

(psu x10−3)<−5

0

>5(b) θ|σ

70S 50S (°C)<−0.05

0

0.05(c) hσ

70S 50S (m)<−200

0

>200

Latitude

Tim

e (y

r)

(d) S|z

70S 50S

200

400

600

800

1000

(psu x10−3)<−5

0

>5

Latitude

(e) θ|z

70S 50S (°C)<−0.05

0

>0.05

Transect location

Longitude

Latit

ude

150E 110W 10W 90E

80S

40S

0

Figure 19: As in Fig. 18, but for the meridional transect shown in the bottom right inset.

68

60W 10W 40E

0

20

40

−0.2

−0.3

0.2

0.4

0.5

Tim

e la

g (y

r)

(a) S|z

60W 10W 40E

0

20

40

−0.3

−0.6−0.7

(b) θ|z

60W 10W 40E

0

20

40

−0.3

−0.2

−0.20.3

0.6

(c) S|σ

Tim

e la

g (y

r)

60W 10W 40E

0

20

40

−0.3

0.3

0.5

−0.2

0.6

(d) θ|σ

60W 10W 40E

0

20

40

−0.3

−0.5−0.7

(e) hσ

Longitude

Tim

e la

g (y

r)

Figure 20: Lagged correlation of Atlantic overturning (ψatl) vs (a, b) S − θ at the meandepth of the σ41.50 isopycnal surface, (c, d) S − θ on σ41.50, and (e) the depth of the σ41.50

surface along the zonal transect shown in Fig. 18. Positive (negative) correlations areshown by solid (dashed) contours. Positive time lags indicate ψatl leading the respectivehydrographic variables.

69