368
in contrast to the case depicted in Fig. 8, the spinup is stopped at a significant distance from the equilibrium point (measured according to the
salinity gradient - the temperature gradient has already equilibrated), and
a switch to a different atmospheric transport model occurs. All models
then further decrease their salinity gradients while moving on their separate T = 0 curves, which continue (with the slope indicated in Fig. 8) to
the left of their focal point, so model temperature gradients diverge from
each other. All steady states are different, and the steady state SST of a
coupled GCM is likely to be further away from the observations than the
SST of the spinup, which is often forced towards climatology.
A spinup may have to be considerably longer than often practiced, if
it is used for flux adjustment and subsequent coupled runs. If the experience of Sausen et al. (1987) can be generalised, even 3000 years are
too short, which is likely to have affected both the 150 year and the 475
year spinup runs discussed by Murphy (1995), and possibly even the experiments of Rahmstorf (1995a). The 30,000 year spinup of Sausen et al.
(1988) virtually eliminated the drift, but this could have been aided by
restoring salinity boundary conditions, as assumed by Rahmstorf (1995a).
In a GCM, a shift in convection patterns might be able to amplify drifts
that would be very small in a purely advective model as the one considered here. It is not possible to point at the single most important cause of
climate drift after flux adjustment, but a very well equilibrated spinup is
a necessary condition for its avoidance.
6 The 'neutrally buoyant mode'
Saravanan and McWilliams (1995) see a mode of variability in T and S
in an idealised coupled model that is nearly compensated in density and
hence not felt by the flow field; they call it the neutrally buoyant mode.
It is associated with the portion of the surface fluxes that is unaffected
by feedbacks; if atmospheric transports change due to the temperature
variability of the neutrally buoyant mode, they are likely to modify T and
S such that density is no longer compensated.
The neutrally buoyant (or neutral) mode manifests itself in the model
used here, and is another contender for causing drifts in a coupled model
that are difficult to control. A low-latitude sinking equilibrium is obtained
with model #5 under the standard parameter set of Table 1. Figure 9 shows
the response of this steady state to a salinity perturbation of 0.2 psu to the
in contrast to the case depicted in Fig. 8, the spinup is stopped at a significant distance from the equilibrium point (measured according to the
salinity gradient - the temperature gradient has already equilibrated), and
a switch to a different atmospheric transport model occurs. All models
then further decrease their salinity gradients while moving on their separate T = 0 curves, which continue (with the slope indicated in Fig. 8) to
the left of their focal point, so model temperature gradients diverge from
each other. All steady states are different, and the steady state SST of a
coupled GCM is likely to be further away from the observations than the
SST of the spinup, which is often forced towards climatology.
A spinup may have to be considerably longer than often practiced, if
it is used for flux adjustment and subsequent coupled runs. If the experience of Sausen et al. (1987) can be generalised, even 3000 years are
too short, which is likely to have affected both the 150 year and the 475
year spinup runs discussed by Murphy (1995), and possibly even the experiments of Rahmstorf (1995a). The 30,000 year spinup of Sausen et al.
(1988) virtually eliminated the drift, but this could have been aided by
restoring salinity boundary conditions, as assumed by Rahmstorf (1995a).
In a GCM, a shift in convection patterns might be able to amplify drifts
that would be very small in a purely advective model as the one considered here. It is not possible to point at the single most important cause of
climate drift after flux adjustment, but a very well equilibrated spinup is
a necessary condition for its avoidance.
6 The 'neutrally buoyant mode'
Saravanan and McWilliams (1995) see a mode of variability in T and S
in an idealised coupled model that is nearly compensated in density and
hence not felt by the flow field; they call it the neutrally buoyant mode.
It is associated with the portion of the surface fluxes that is unaffected
by feedbacks; if atmospheric transports change due to the temperature
variability of the neutrally buoyant mode, they are likely to modify T and
S such that density is no longer compensated.
The neutrally buoyant (or neutral) mode manifests itself in the model
used here, and is another contender for causing drifts in a coupled model
that are difficult to control. A low-latitude sinking equilibrium is obtained
with model #5 under the standard parameter set of Table 1. Figure 9 shows
the response of this steady state to a salinity perturbation of 0.2 psu to the