Evaporation or precipitation serves to concentrate or dilute the dissolved
salt content of surface sea water, and thus directly affects its density.
The haline density flux 
is given by [ Stern, 1975]:
where 
, is the haline
contraction coefficient, and 
and 
are the densities of
pure and sea water. The haline contribution to the buoyancy flux is often
comparable to that of the thermal buoyancy flux 
, where Q
is the net heat flux and 
is
the thermal expansion coefficient).
Schmitt, et al. [1989] compared the annual mean thermal and haline buoyancy flux
estimates for the North Atlantic by contouring the absolute value of the
ratio of the two components. They find that in the subtropical gyre the
haline density flux is often larger than the thermal density flux, though
the general pattern is one of thermal flux dominance in high and low
latitudes. However, this is based entirely on open ocean estimates of
surface fluxes. Continental runoff and ice freezing and melting processes
contribute to significant concentration of the haline density flux at the
coasts, where it can easily dominate the buoyancy change locally.
The role of the haline buoyancy flux in ocean dynamics was first examined by Stommel [1961] with a simple box model. He found multiple stable states of the system, which was forced by both heating/cooling and evaporation/precipitation. Other box models were discussed by Rooth [1982] and Walin [1985]. Bryan [1986] showed how a general circulation model would display instability of the thermohaline circulation when restoring boundary conditions were replaced by flux boundary conditions for the salinity. He coined the phrase ``haline catastrophe'' to refer to the freshening of high latitude surface waters and shutdown of the thermohaline circulation. Since then a great variety of models of varying complexity have focussed on the sensitivity of the thermohaline circulation to freshwater forcing [ Marotske, et al. 1988; Huang, et al., 1992; Huang and Chou, 1994; Quon and Ghil, 1992; Zhang, et al., 1993; Weaver and Sarachik, 1991; Weaver et al., 1991; Marotske and Willebrand, 1991; Stocker and Wright, 1991; Wang and Birchfield, 1992; Weaver and Hughes, 1994; Tziperman, et al., 1994; Mikolajewicz and Maier-Reimer, 1994]. Holland, et al. [1995] discuss such models in their review. The essential point is that the ``conveyor belt'' of the thermohaline circulation can be very sensitive to the freshwater flux at high latitudes. Many of these models display fluctuations on millenial time scales. However, Delworth, et al. [1993] have carefully examined the behavior of a realistic circulation model, and found significant variability on 40--60 year time scales. Advection of heat and salt play an important role in the variability, which appears to be governed by the strength of a baroclinic gyre set up in mid-basin. The persistence of sea-surface temperature anomalies is an important element, so the common relaxation boundary conditions, which couple the ocean too tightly to the atmosphere, are not appropriate. Huang [1993] has pointed out the importance of real freshwater flux as a boundary condition for models, in part because of the Goldsbrough-Stommel circulation, which is missing when a psuedo salt flux is applied. Also of interest is the modelling study of Shaffer and Bendtsen [1994], who find a great sensitivity of the thermohaline circulation to the transport through Bering Strait. Because this supplies freshwater to the Atlantic, it acts as a negative feedback on the thermohaline conveyor belt. They propose that a more variable climate would be realized with a higher sea level and thus greater Bering Strait transport. Such conditions may have been achieved during the last interglacial period, when the Greenland Ice Cores indicate that the climate had less stability than at present Dansgaard, et al., 1993]. Other paleoclimate records suggest that sudden changes in the North Atlantic thermohaline circulation occured many times during the last deglaciation [ Lehman and Keigwin, 1992], when substantial discharges of meltwater are thought to have occurred at high latitudes. We also note that strong haloclines in the tropics, formed by the rainfall under the ITCZ, may also have an impact on upper ocean mixing and heat exchange processes there [ Lukas and Lindstrom, 1991; Carton, 1991; Delcroix and Hénin, 1991].
There is also historical evidence that perturbations of the hydrologic cycle do influence the climate system. In particular, fresh salinity anomalies in the subpolar basin have been observed to survive for the order of a decade while being advected by the surface gyre. The Great Salinity Anomaly (GSA) of the 60's--80's was observed for 15 years as it traveled around the entire subpolar North Atlantic gyre [ Dickson,et al., 1988]. There is strong evidence that the GSA had substantially supressed deep water formation in the Greenland and Norwegian Sea and flux in the western boundary currents [ Schlosser, et al., 1991]. The GSA is thought to have arisen from excess ice discharge from the Arctic through Fram Strait. An increase in the annual export of ice of only 20% [ Aagaard and Carmack, 1989] could have provided the necessary freshwater anomaly.
A potentially related process has been identified by Deser and Blackmun [1993]. They found a mode of Atlantic surface and air temperature variability with a dipole structure concentrated in the western North Atlantic, by examining 90 years of temperature data. Winter temperatures east of Newfoundland are out-of-phase with those off the US mid-Atlantic states. The mode has an amplitude of about 1 C and a 10--12 year period. Interestingly, it appears to be related to the ice extent in the Labrador sea, with Labrador sea ice leading low temperatures off Newfoundland by one-two years. Advection of a low salinity cap formed by the melting ice would be consistent with such a relation.
Thus, there is substantial evidence that the thermohaline circulation has intimate ties to climate, especially on the decadal time scale. It is therefore important to further define its sensitivity to perturbations in the freshwater flux, which is only hinted-at in the currently meager data sets. Both improved models and an expanded observational base are required.