The laboratory experiments on brine pocket migration were on a scale that is about one-tenth that of our in situ experiment. This small scale might remove some of the larger brine features, and not allow the feeding of brine pockets into brine channels that is observed in sea ice (Niedrauer and Martin, 1979; Weeks and Ackley, 1986). Gravity drainage of brine is associated with brine tubes or slots, open at the bottom to sea water (Lake and Lewis, 1970; Weeks and Ackley, 1986). Observations of fluid and salt fluxes in brine channels (Niedrauer and Martin, 1979) in sea ice suggest that the flow rates and frequency of occurrence of these channels are high enough to produce significant heat and salt fluxes. Our data shows evidence of the transient effects of brine tubes on temperatures. Fig. 7 is a closer view of temperatures from five adjacent thermistors near the bottom of the ice, showing evidence of a brine tube that extends about 300mm into the ice from the sea, lowering and raising temperatures locally.
Cox and Weeks (1981) parameterize gravity drainage from sea ice based
on temperature gradients. However, in their model, gravity drainage
stops when the brine volumes fall below 50 
, which corresponds to
temperatures below -5 
C. Our data analysis excludes
temperatures above -5 C, so that gravity drainage would not be
significant according to their model. Brine expulsion remains as a
likely mechanism -- however, we do observe evidence of nearby brine
tubes at temperatures below -5 C, as noted above.
Here we consider convection in brine tubes, closed at both ends. For a vertical cylinder of fluid heated from below the Rayleigh number is (Niedrauer and Martin, 1979)
where 
is brine density, D is the molecular diffusivity of
salt, 
is the dynamic viscosity of the brine, g is the
gravitational acceleration and r is the radius of the cylinder. The
critical value of Ra for the onset of convection is 68. If Ra is
below this value, there is no convection, and diffusion of salt is the
dominant mechanism. Above this value, convection begins, driven by the
buoyancy differences generated by the salinity gradient. We calculate
that for the properties of brine at -10 C and a temperature
gradient of 10 C m 
, the critical Rayleigh number is
exceeded when the diameter of the brine tube is greater than 1mm. This
is larger than most brine pockets.
Many brine tubes are tilted at an angle to the vertical (Niedrauer and
Martin, 1979), which alters their convective behaviour. Criminale and
Lelong (1984) perform a stability analysis of tilted brine drainage
channels, and find two optimum angles of the channel for convection or
overturn driven by brine density gradients, one being vertical and the
other being at about 45 . This suggests that there is a natural
preference for either vertical or 45 tubes, since at these
angles the instability driving convection is strongest. All of the
above analyses predict no convection in the brine tubes at small
temperature gradients.
However, Woods and Linz (1992) have shown that a slot or fracture that is inclined at some angle to the vertical, containing fluid of different thermal conductivity to the surrounding rock, will convect at any Rayleigh number, even if the thermal gradient in the rock is stabilizing, due to tilted isotherms. They find an exact solution, which becomes singular as a critical Rayleigh number is approached. Vertical temperature gradients impose brine density gradients in brine tubes, a case covered in the work of Woods and Linz. However, their analysis (and that of Crimimale and Lelong, 1984) does not include the important effect of the phase change at the tube face, or allow for (or predict) the movement of the tube through the ice. Nevertheless, the mechanism driving the instability, that of tilted isotherms and isopycnals due to the different thermal conductivities of the brine and the ice, is present in sea ice, and we expect qualitatively similar results. That is, there would be no critical Rayleigh number for the onset of convection in titled brine tubes, and convection would be observed at all gradients and in any tube no matter how narrow.