observed wind stress anomalies are often added
to the model’s own mean wind stress climatology in the initialization phase. Similar considerations apply if ocean data such as XBT profiles or
satellite SSTs are assimilated during the initialization. There is little experience here on decadal
or longer time scales (for example hindcasting of
North Atlantic Oscillation variability).
4 Forced climate change. Similar considerations
apply as in 2 and 3 above. At the stage where
any forcing is applied to the model, the ‘control’
(unforced) run will ideally not contain any
climate drift. At the very least the rate of drift
needs to be smaller than the rate of climate
change due to the forcing. Again, the methods
of Section 2.3.4.2 are used in practice.
2.3.4.2 Practical techniques
For applications 2 and 4 above, the primary
requirement of a model initialization procedure is
that the model drift should be acceptably small
(small compared with the signals that are being
studied). A wide variety of methods have been used
(see Stouffer and Dixon, 1998, for a review). However, the methods can be split into two classes:
a Initialize the model from an estimate of the present climate (particularly ocean) state, and simply run the model forward in time until the drift
is considered acceptably small (e.g. Guilyardi
and Madec, 1997; Johns et al., 1997b; Barthelet
et al., 1998; Gordon et al., 2000). This has the
advantage of being conceptually and technically
simple, and by studying the processes by which
the model adjusts from its (realistic) initial state
to its own preferred climatology, insight can be
gained into the model’s systematic errors.
b Use separate spin-up integrations of the atmospheric and oceanic components to bring the
ocean model as close as possible to a quasi-equilibrium state (no long-term drift), followed by a
final recoupled phase (e.g. Manabe et al., 1991;
Bryan, 1998). The acceleration technique of
Bryan (1984) can often be used to reduce the
computing requirement of the ocean-only phase.
This technique has the advantage that it may
result in a smaller residual drift than method a
(though this is not guaranteed because important atmospheric feedbacks may be missing
from the ocean-only phase, leading to large
drifts when the model is recoupled).
Either method has been used with both fluxadjusted and non-flux-adjusted models. However,
for flux-adjusted models experience has shown
that a long (multicentury) ‘calibration’ run is
required to define the flux-adjustment terms (see
Section 2.3.3), whereas shorter spin-up runs may
be sufficient in the non-flux-adjusted case.
2.3.5 Coupled model simulation of
present and past climates
2.3.5.1 Sea surface temperature
The ocean influences atmospheric climate largely
through the SST, so for climate studies, those ocean
processes that influence SST on the time scale of
interest are the most important. Thus it may be
that for, say, study of interannual climate variability a model with a poor representation of North
Pacific Deep Water may do an adequate job. However, on the decadal to century time scales for
which global coupled models are most widely
used, a large number of ocean processes are potentially important. Some of these are discussed in
this section.
Because of the importance of SST much attention is given by coupled modellers to the SST field
simulated by the models. As discussed in Section
2.3.3 and (Fig. 2.3.2, see Plate 2.3.2, p. 76), the
choice of a suitable flux-adjustment field enables
any model to produce a realistic SST field, at the
cost of unphysical terms (often large) in the global
heat balance. The SST error field from the
HadCM3 model (Fig. 2.3.2a, see Plate 2.3.2a, p. 76)
shows an error pattern that appears to be common
among non-flux-adjusted models: the cool errors
in the North and Equatorial Pacific, and the warm
errors in the Southern Ocean and in the eastern
tropical Atlantic and Pacific, can also be seen
(with various amplitudes) in the NCAR CSM
(Boville and Gent, 1998) and in the OPA/ARPEGE
model (Madec and Delecluse, 1997), and have
been the subject of much diagnostic effort.
2.3.5.2 Heat and freshwater transports
The importance of a good representation of the
large-scale heat transports through the atmosphere–ocean system was discussed in Section 2.3.3.
More detailed comparison of ocean model heat
transports with estimates from hydrographic
sections has also given useful insights into model
performance. For example, Banks (2000) compares
2.3 Coupled Ocean–Atmosphere Models
87
Wood and Bryan
to the model’s own mean wind stress climatology in the initialization phase. Similar considerations apply if ocean data such as XBT profiles or
satellite SSTs are assimilated during the initialization. There is little experience here on decadal
or longer time scales (for example hindcasting of
North Atlantic Oscillation variability).
4 Forced climate change. Similar considerations
apply as in 2 and 3 above. At the stage where
any forcing is applied to the model, the ‘control’
(unforced) run will ideally not contain any
climate drift. At the very least the rate of drift
needs to be smaller than the rate of climate
change due to the forcing. Again, the methods
of Section 2.3.4.2 are used in practice.
2.3.4.2 Practical techniques
For applications 2 and 4 above, the primary
requirement of a model initialization procedure is
that the model drift should be acceptably small
(small compared with the signals that are being
studied). A wide variety of methods have been used
(see Stouffer and Dixon, 1998, for a review). However, the methods can be split into two classes:
a Initialize the model from an estimate of the present climate (particularly ocean) state, and simply run the model forward in time until the drift
is considered acceptably small (e.g. Guilyardi
and Madec, 1997; Johns et al., 1997b; Barthelet
et al., 1998; Gordon et al., 2000). This has the
advantage of being conceptually and technically
simple, and by studying the processes by which
the model adjusts from its (realistic) initial state
to its own preferred climatology, insight can be
gained into the model’s systematic errors.
b Use separate spin-up integrations of the atmospheric and oceanic components to bring the
ocean model as close as possible to a quasi-equilibrium state (no long-term drift), followed by a
final recoupled phase (e.g. Manabe et al., 1991;
Bryan, 1998). The acceleration technique of
Bryan (1984) can often be used to reduce the
computing requirement of the ocean-only phase.
This technique has the advantage that it may
result in a smaller residual drift than method a
(though this is not guaranteed because important atmospheric feedbacks may be missing
from the ocean-only phase, leading to large
drifts when the model is recoupled).
Either method has been used with both fluxadjusted and non-flux-adjusted models. However,
for flux-adjusted models experience has shown
that a long (multicentury) ‘calibration’ run is
required to define the flux-adjustment terms (see
Section 2.3.3), whereas shorter spin-up runs may
be sufficient in the non-flux-adjusted case.
2.3.5 Coupled model simulation of
present and past climates
2.3.5.1 Sea surface temperature
The ocean influences atmospheric climate largely
through the SST, so for climate studies, those ocean
processes that influence SST on the time scale of
interest are the most important. Thus it may be
that for, say, study of interannual climate variability a model with a poor representation of North
Pacific Deep Water may do an adequate job. However, on the decadal to century time scales for
which global coupled models are most widely
used, a large number of ocean processes are potentially important. Some of these are discussed in
this section.
Because of the importance of SST much attention is given by coupled modellers to the SST field
simulated by the models. As discussed in Section
2.3.3 and (Fig. 2.3.2, see Plate 2.3.2, p. 76), the
choice of a suitable flux-adjustment field enables
any model to produce a realistic SST field, at the
cost of unphysical terms (often large) in the global
heat balance. The SST error field from the
HadCM3 model (Fig. 2.3.2a, see Plate 2.3.2a, p. 76)
shows an error pattern that appears to be common
among non-flux-adjusted models: the cool errors
in the North and Equatorial Pacific, and the warm
errors in the Southern Ocean and in the eastern
tropical Atlantic and Pacific, can also be seen
(with various amplitudes) in the NCAR CSM
(Boville and Gent, 1998) and in the OPA/ARPEGE
model (Madec and Delecluse, 1997), and have
been the subject of much diagnostic effort.
2.3.5.2 Heat and freshwater transports
The importance of a good representation of the
large-scale heat transports through the atmosphere–ocean system was discussed in Section 2.3.3.
More detailed comparison of ocean model heat
transports with estimates from hydrographic
sections has also given useful insights into model
performance. For example, Banks (2000) compares
2.3 Coupled Ocean–Atmosphere Models
87
Wood and Bryan