The Weddell Sea sector of the Southern Ocean is a region of intense air-sea-ice interaction during winter, stirred by frequent, fast moving cyclones that mix enough heat from the underlying ocean to keep the ice cover thin, despite its high latitude and the proximity to the Antarctic continent. Processes by which heat, salt, and momentum are exchanged in the lower atmosphere/sea ice/upper ocean system were studied thoroughly during the 1994 Antarctic Zone Flux Experiment (ANZFLUX), with turbulent fluxes measured directly in the upper ocean from two separate drift stations (one over Maud Rise, the other in the so-called Warm Regime about 300 km SW of Maud Rise) during several storms (McPhee et al., 1996; 1999). In addition, unmanned buoys left when the first drift station was abandoned provided a record of ice growth and temperature gradient, and upper ocean properties for the remainder of the season (Muench et al., 2001; McPhee et al., 1999).

From ANZFLUX and other studies (Martinson 1990; Wadhams et al. 1987), it appears that the ice cover in the central and eastern Weddell Sea grows rapidly early in the winter, but rather quickly reaches a thickness (40-60 cm) where heat conduction along the thermal gradient in the ice matches heat flux into the surface mixed layer (SML) from the underlying, relatively warm WDW. This sounds simple, but is not. It involves a dynamic sequence that goes something like this: during the calm between storms, air temperatures and the upper surface radiation balance are such that the thin ice, as well as any open or surface water (e.g. from flooding induced by snow load and stress) freeze rapidly. As an approaching storm begins moving the ice, negative buoyancy flux associated with freezing enhances shear-driven turbulence and there is often intense mixing at the interface between the SML and the WDW, typically a hundred meters below the surface. This mixes heat upward from the pycnocline which is dispersed rapidly through the boundary layer, and as the storm intensifies, upward sensible heat flux at the ice/ocean interface exceeds upward conduction in the ice (values as high as 300 W m-2 were observed near the surface during ANZFLUX storms [McPhee 1999]), and melting replaces growth. Eventually, the stabilizing buoyancy flux accompanying the meltwater input reduces turbulence levels and curtails upward heat entrainment. As the storm abates, ocean heat flux at the interface diminishes until the heat balance is again dominated by upward conduction in the ice, and the growth part of the cycle begins again. This conceptual picture, with latent heat at the ice/ocean interface serving as a “thermal flywheel,” is further complicated by surface flooding associated with snow load and ice stress, and by large excursions in the pycnocline below the mixed layer (“ocean weather”).

The water column in the central and eastern Weddell is often only marginally stable (e.g., Martinson and Ianuzzi 1998), and there are many regions where a combination of modest ice growth and heat loss to the atmosphere could drive the system to convective instability, i.e., where destabilizing surface buoyancy flux would readily mix water from the pycnocline into the well mixed upper layer. But this surface driven mixing is difficult to maintain, because as soon as enough heat is mixed from below to warm the mixed layer, melting produces the buoyancy necessary to re-establish bulk stability. The thermal barrier (Martinson 1990) associated with the phase change at the ice/ocean interface exerts a powerful negative feedback to convection. Although the ice cover at the end of the ANZFLUX Maud Rise drift was only 35-40 cm thick, it held enough buoyancy to seemingly prevent its wholesale destruction over a wide area. Yet soon after our departure in August, 1994, satellite imagery showed a widespread area of opening in the ice pack just northeast of our Maud Rise drift (Drinkwater 1997). Clearly, our understanding of the dynamics and thermodynamics of the ice/upper ocean system was incomplete.

The key to understanding how the stabilizing feedback from ice melt can be overcome may lie with nonlinearities in the equation of state (NES) for seawater. Akitomo (1999a;b) described a mechanism for sustaining deep convection, even when there is stabilizing surface buoyancy flux from melting. The thermal expansion factor for seawater (?) increases with pressure so that if a layer of cold, relatively fresh water overlies a layer of warmer, saltier water, the relative stability of the two layers will change if the depth of the interface changes (an animated illustration is provided at the website mil/thermobaricity). Under certain conditions, this phenomenon (thermobaricity, a term coined by T. McDougall) can lead to an instability with overturning and mixing. Akitomo (1999a) pointed out the distinction between convection driven by destabilizing surface buoyancy flux (augmented at depth by thermobaric effects), which he termed Type I convection, and two-layer Type II convection, capable of maintaining intense mixing regardless of surface buoyancy flux. He demonstrated that much of the Weddell Sea is potentially vulnerable to Type II thermobaric instability, in contrast to the Greenland Sea. The crucial point is that once triggered, Type II thermobaric overturn no longer relies on being driven by surface buoyancy and momentum flux, but instead derives its mixing energy from the potential energy of the water column.

McPhee (2000) extended Akitomo’s analysis to derive a quantitative measure of the heat loss required to drive a given thermohaline upper ocean structure to thermobaric instability. He showed with simple one-dimensional modeling that a significant fraction of the T/S profiles observed over Maud Rise in 1994 were indeed susceptible to thermobaric instability by late winter, hence that thermobaricity may have been a major factor in the observed opening. From the 1994 study, the least thermobarically stable profiles had in common a fairly homogeneous layer of relatively warm, more saline water below the surface mixedlayer. From measurements that we made as the station drifted across one such feature, active mixing was found within one such subsurface “step” layer. We inferred that its most likely source was cabbeling instability, i.e., the dependence of ? on T (Fofonoff 1956). Cabbeling is usually associated with isopycnal mixing, e.g. at ocean fronts, but it is also possible for cabbeling to occur in the vertical, when stratification is weak (Foster 1972). The instability may be initiated by increased density of the surface mixed layer by salt rejection during ice growth. In the case of ANZFLUX observations, however, a more likely scenario involves the differential advection by Ekman transport of mixed-layer water with slightly enriched salinity over a filament of WDW uplifted along the eastern flank of Maud Rise. The flank is a preferred site for this mechanism, since the surface layer salinity over the top of Maud Rise is slightly higher than the surroundings. As with seamounts elsewhere, the slopes of Maud Rise appear to be energetic, with significant tidal and small-scale eddy activity. In addition, there is a water mass boundary encircling the feature that separates the Taylor column structure over Maud Rise (with significantly lower Tmax values) from the surrounding Weddell Gyre water (Muench et al. 2001). McPhee’s (2000) assessment of thermobaric stability from the 1994 experiment suggested that the least stable profiles were found between the 2500 and 3000 m isobaths of Maud Rise. Analysis of other winter profiles, most notably from the 1986 Polarstern section, supports this view (McPhee 2003).

Following recommendations from a workshop devoted to thermobaricity and other NES issues held in Monterey, CA, in October, 2000, a modeling effort was undertaken to study Weddell Sea thermohaline structure and forcing, using a Large Eddy Simulation (LES) numerical model (Moeng 1984; Harcourt et al. 2002). The model was initialized to the least thermobarically stable profile discussed by McPhee (2000), and forced with surface stress and conductive heat flux in the ice cover inferred from buoy measurements some distance away. Fig. 2 (Harcourt, in preparation) summarizes some model results from this current effort. The LES reveals a complex instability that combines cabbeling instability at the bottom of the pycnocline with thermobaric convection below the SML. Following three simulated days of ice formation over a 100 m thick SML, a distinct and decoupled Internal Mixed Layer (IML) with O(1cm s-1) rms velocity scales forms below the thermocline, entraining upwards 30 m into the SML and downwards more than 300 m. Initially, mixing within the lower thermocline produces a mean density inversion due to cabbeling between SML fluid and warm, salty deep water.

The vertical flux of buoyant parcels out of this density inversion produces enough Turbulent Kinetic Energy (TKE) to create a relatively homogeneous layer delimited above and below by the neutral buoyancy depths of these parcels. As this layer grows, this neutral depth for relatively cold parcels (Fig. 2a) accelerated downward from the cabbeling instability is increasingly magnified by thermobaricity, and within hours TKE production within the IML becomes dominated by the thermobaric component of buoyancy flux (Fig. 2c). The LES modeled IML bears a striking resemblance (Fig. 2b) to profiles observed near this location during ANZFLUX. Like Type II thermobaric mixing, this combination cabbeling and thermobaricity also raises heat towards the surface without as much surface warming and associated ice melt as SML deepening through entrainment. This instability might be called thermobaric cabbeling or Type III thermobaricity to distinguish it from others. Further analysis shows that conditions for the same instability can also be approached either by vertically stretching the water column or through differential advection of the SML over deep water, in addition to the approach modeled here where SML density first increases due to surface ice formation.

Because of the thermal barrier effect, Type II instability is most effective when the greatest temperature contrast occurs over the shortest vertical distance, i.e., well defined steps with the large T/S differences, but vanishing difference in density at the pressure surface separating the two layers (McPhee 2000). It appears that the Type III process could play a key role in preconditioning a particular region for convection strong enough to destroy the ice cover. Several other mechanisms may also contribute to mixing in the interior of the water column, among them (i) double diffusive convection (“DDC”) (Muench et al., 1990); (ii) internal wave breaking (e.g., Gregg, 1987); and (iii) differential advection of a horizontal density gradient in a shear flow (Crawford et al. 1999). However, for removing the ice cover by purely vertical processes, the prime test of any combination of factors is whether the mixing of heat upward can be sustained long enough to overcome the strongly stabilizing influence of ice melt, i.e., the thermal barrier. By a simple thought experiment involving the deepening of the interface between two layers starting at neutral thermobaric stability, McPhee (2003) showed that the rate of heat transfer at the ice/ocean boundary is another critical factor in the competition between buoyancy added by melting, and that subtracted by conversion of potential energy of the water column.

The small-scale processes described above rely for their preconditioning on the larger scale circulation. Satellite imagery (Fig. 4) and the animation of the movement of the 1994 polynya clockwise around the north west flanks of Maud Rise (http:// /mrd /Images /Polynya_movie.gif) indicate the controlling role played by Maud Rise. Thus, to understand the preconditioning requires knowledge of the circulation regional to the rise.

Northeast of Maud Rise, waters from the Antarctic Circumpolar Current are diverted southwards into a confused region of eddies and mixing (Gouretski & Danilov 1993). From this region, a mean flow of WDW, a several hundred meters thick layer characterized (at Maud Rise) by a temperature maximum at ~0.6 deg C, 150m depth, salinity >34.67 psu, heads southwestward towards Maud Rise, carrying with it various eddies. The warm WDW is an obvious source of heat for polynya formation, yet is usually insulated from the sea ice by the intervening cooler, fresher Antarctic Surface Waters. The scale of Maud Rise (vertically a rise to ~1600 m from a seafloor of ~5000 m, with a horizontal diameter of ~200 km, much greater than the internal Rossby radius of ~5 km) and the weak stratification imply a trapped “Taylor column” circulation pattern (e.g. Ou 1991; Alverson & Owens 1996), typified by a cap of dense water overlying the crest of the rise. A dense cap (and the corresponding “warm halo” around the rise) is observed in the several years of hydrographic data (e.g. Muench et al. 2001). A year of moored current measurements (Bersch et al. 1992) show flows consistent in direction with a Taylor circulation; strongly coherent in the vertical; and O(5 cm s-1), suggesting a NE-SW transit time of a few months. Theory and modeling (Ou 1991; Holland 2001a) indicate also vertical displacements of isopycnals on the flanks and outside the region of the rise, and accompanying circulation cells both over the rise and trapped to its sides. The mean ice drift (Kottmeier and Sellmann 1996) is climatologically northeastward, opposed to the ocean mean circulation, and O(5 cm s-1), although storms can induce O(1) m s-1 motion (McPhee et al. 1996). The mean 10-m winds have a minimum in this region (Kottmeier and Sellmann, 1996). Within this hydrographic regime, several mechanisms could create conditions favorable for the onset of small-scale NES convective processes. Some require lateral displacement of a portion of the dense cap waters off the rise, for example by Ekman transport of the surface layer waters (c.f. the hypothesis of Muench et al. [2001] that the 1994 polynya formation following a sudden storm) or by eddy shedding. Transient quasi-geostrophic flow can shed large (>100 km), cyclonic circulations (Huppert and Bryan 1977; Verron 1984) which may propagate downstream with the mean gyre circulation, c.f, the westward progression of the 1970s Weddell Polynya (Martinson et al. 1981). Another obvious candidate is the mixed layer thinning and the WDW shoaling associated with the Taylor circulation induced isopyncal displacements (Holland 2001a). Further factors include ice divergence/convergence, changes in wind stress, passage of storms, vertical stretching of the water column, variability in depth and temperature of the core of WDW impinging on the Rise, and the influence of eddies that are likely to be frequently embedded in the WDW inflow. Studies incorporate some of these processes in simulations of polynya formation (e.g., Holland 2001b), but in general we anticipate that a combination of processes is necessary to successfully precondition the water column for deep convection and polynya formation and explain the longevity of the deep convective patch once it is initiated.