A global version of the GFDL modular ocean model is forced using conve
ntional restoring boundary conditions (BCs), mixed BCs (i.e., restorin
g the upper-level temperature but specifying a fixed salt flux), and s
tochastic fluxes of both heat and freshwater. The climatology of the m
odel is found to drift if stochastic freshwater fluxes are applied at
high latitudes under mixed BCs. The drift is global in extent: the oce
an is generally warmer in the North Pacific and Weddell Sea but cooler
and fresher at depths elsewhere in the Southern Ocean and in the Nort
h Atlantic. There is a slight reduction (by about 5%) in the meridiona
l overturning of the Southern Ocean and the North Atlantic. The drift
of the barotropic flow is most pronounced in the Southern Ocean and is
associated with a permanent meandering of the Antarctic Circumpolar C
urrent. The drift occurs within a few decades, suggesting that it may
be important in enhanced greenhouse scenarios for early next century t
hat have been obtained using coupled atmosphere-ocean GCMs. It is also
possible that some of the intrinsic variability identified in the sam
e models is actually a residual drift.The drift depends upon convectiv
e adjustment to occur but can be amplified by the surface heat flux pa
rameterization, both locally and by an additional feedback associated
with large-scale flow changes. In an extreme case, the latter leads to
a total collapse of the thermohaline circulation associated with Nort
h Atlantic Deep Water Formation. A similar mechanism underlies the dri
ft that can occur when the switch from restoring to mixed BCs is made.
The heat flux feedback represents the atmosphere-ocean coupling in th
e model, so this aspect of the drift can be regarded as a coupled mode
that actually contributes to the mean state of the coupled system. Th
e existence of such modes makes some climatic drift in coupled models
inevitable, if the individual components are equilibrated separately p
rior to coupling. The applicability of these results to more sophistic
ated coupled models depends, in part, upon how well the restoring BC o
n temperature captures the heat flux feedback they exhibit.