GCMs explicitly include parameterizations of
clouds, precipitation, surface evaporation and soil
moisture accumulation. However, the runoff is
often transported instantaneously from the accumulation region to its river outflow point. This
can lead to errors in the seasonal timing of the
freshwater input to the ocean, and schemes are
now being developed that take account of the time
delay from the river catchment to the outflow
(Hagemann and Dümenil, 1998). Iceberg calving is
modelled at best as a fixed water flux into the
ocean (e.g. Gordon et al., 2000). Since the ice
sheet dynamics by which the iceberg flux responds
to variations in snow accumulation and surface
temperature are not well understood, a more
sophisticated parameterization seems inappropriate at this stage.
The water budget of marginal seas such as the
Mediterranean, which may be enclosed at the
model grid scale, presents a particular challenge to
coupled models. Unlike the heat budget, there is
no strong atmospheric feedback to stabilize the
salinity if a small water imbalance causes it to drift
from realistic values. Such imbalances can accumulate over a period to give very large salinity
errors. Most models allow some form of exchange
between the Mediterranean and the Atlantic,
either through a simple mixing parameterization
(e.g. Manabe et al., 1991; Gordon et al., 2000) or
by opening an artificially wide ‘Gibraltar Straits’
in the model grid (e.g. Gent et al., 1998; Voss
et al., 1998), but seas with no link to the ocean
(e.g. the Caspian Sea, whose water source is runoff
and whose sink is evaporation) present a particular challenge.
An additional technical issue arises in the specification of the surface boundary condition for
fresh water. Most ocean GCMs in current use,
both rigid lid (Bryan, 1969) and ‘dynamic free surface’ (Killworth et al., 1991), are formulated to
conserve ocean volume rather than mass, and
require the natural boundary condition on fresh
water to be converted to a boundary condition
that represents a virtual salt flux (Huang, 1993).
The response of the surface salinity to a surface
freshwater flux Q s (in m s
91
) is
dS/dt:9(SQ s )/d
(2.3.1)
where Q s is the water flux into the ocean and d is
the depth of the mixed layer (or top model grid
box). Where there is net precipitation (Q90), the
sea surface rises, resulting in lower salinity, and
where there is net evaporation (Q:0), the surface
falls, resulting in higher salinity. However, in a
rigid lid model the surface is fixed (d is constant).
If (2.3.1) is applied, the model’s local salinity
responds correctly to the flux Q s , but because
globally S and Q s are anticorrelated (high salinity
tends to occur in regions of net evaporation, and
vice versa), the global mean salinity will tend to
increase when the model is given a Q s field whose
global mean is zero. This effect has been found to
be significant in practice, and can only be ‘cured’
by replacing S on the right-hand side of (2.3.1)
with a fixed reference salinity S 0 , at the cost of a
local inaccuracy in the salinity response (Gordon
et al., 2000). Formulations of the natural boundary condition are available for both rigid lid
(Huang, 1993) and free surface (Roullet and
Madec, 2000) models; however, there is little
experience to date with their use in long coupled
model runs.
The careful accounting of water discussed in
this subsection is of special importance for nonflux-adjusted models (see Section 2.3.3 below).
Where flux adjustments are used, they are usually
chosen in such a way as to compensate for the
slight imbalances described above.
2.3.2.7 Forcing
The issue of forcing of coupled models is less
complex than that of forcing ocean-only GCMs,
because the climate system as a whole is more selfcontained than the ocean. The following fields typically need to be prescribed for a coupled model:
top of the atmosphere incoming solar radiation,
atmospheric composition of those radiatively
active species that are not explicitly calculated in
the model (see Section 2.3.2.1), and distribution of
vegetation types (see Section 2.3.2.2). Most of
these quantities are relatively well observed for the
present climate state, and estimates exist for past
climates, allowing relatively easy interpretation of
model simulations of a particular climate state
(but see Section 2.3.4 below). However, for longtime scale climate variability and change there is a
potential for feedbacks of the biogeochemical
cycles on climate (Charlson et al., 1987; IPCC,
1995 (Chapter 9); Maier-Reimer et al., 1996;
Levis et al., 1999), and development of models
that include such feedbacks is still at an early stage
(Cox et al., 2000).
2.3 Coupled Ocean–Atmosphere Models
83
Wood and Bryan
clouds, precipitation, surface evaporation and soil
moisture accumulation. However, the runoff is
often transported instantaneously from the accumulation region to its river outflow point. This
can lead to errors in the seasonal timing of the
freshwater input to the ocean, and schemes are
now being developed that take account of the time
delay from the river catchment to the outflow
(Hagemann and Dümenil, 1998). Iceberg calving is
modelled at best as a fixed water flux into the
ocean (e.g. Gordon et al., 2000). Since the ice
sheet dynamics by which the iceberg flux responds
to variations in snow accumulation and surface
temperature are not well understood, a more
sophisticated parameterization seems inappropriate at this stage.
The water budget of marginal seas such as the
Mediterranean, which may be enclosed at the
model grid scale, presents a particular challenge to
coupled models. Unlike the heat budget, there is
no strong atmospheric feedback to stabilize the
salinity if a small water imbalance causes it to drift
from realistic values. Such imbalances can accumulate over a period to give very large salinity
errors. Most models allow some form of exchange
between the Mediterranean and the Atlantic,
either through a simple mixing parameterization
(e.g. Manabe et al., 1991; Gordon et al., 2000) or
by opening an artificially wide ‘Gibraltar Straits’
in the model grid (e.g. Gent et al., 1998; Voss
et al., 1998), but seas with no link to the ocean
(e.g. the Caspian Sea, whose water source is runoff
and whose sink is evaporation) present a particular challenge.
An additional technical issue arises in the specification of the surface boundary condition for
fresh water. Most ocean GCMs in current use,
both rigid lid (Bryan, 1969) and ‘dynamic free surface’ (Killworth et al., 1991), are formulated to
conserve ocean volume rather than mass, and
require the natural boundary condition on fresh
water to be converted to a boundary condition
that represents a virtual salt flux (Huang, 1993).
The response of the surface salinity to a surface
freshwater flux Q s (in m s
91
) is
dS/dt:9(SQ s )/d
(2.3.1)
where Q s is the water flux into the ocean and d is
the depth of the mixed layer (or top model grid
box). Where there is net precipitation (Q90), the
sea surface rises, resulting in lower salinity, and
where there is net evaporation (Q:0), the surface
falls, resulting in higher salinity. However, in a
rigid lid model the surface is fixed (d is constant).
If (2.3.1) is applied, the model’s local salinity
responds correctly to the flux Q s , but because
globally S and Q s are anticorrelated (high salinity
tends to occur in regions of net evaporation, and
vice versa), the global mean salinity will tend to
increase when the model is given a Q s field whose
global mean is zero. This effect has been found to
be significant in practice, and can only be ‘cured’
by replacing S on the right-hand side of (2.3.1)
with a fixed reference salinity S 0 , at the cost of a
local inaccuracy in the salinity response (Gordon
et al., 2000). Formulations of the natural boundary condition are available for both rigid lid
(Huang, 1993) and free surface (Roullet and
Madec, 2000) models; however, there is little
experience to date with their use in long coupled
model runs.
The careful accounting of water discussed in
this subsection is of special importance for nonflux-adjusted models (see Section 2.3.3 below).
Where flux adjustments are used, they are usually
chosen in such a way as to compensate for the
slight imbalances described above.
2.3.2.7 Forcing
The issue of forcing of coupled models is less
complex than that of forcing ocean-only GCMs,
because the climate system as a whole is more selfcontained than the ocean. The following fields typically need to be prescribed for a coupled model:
top of the atmosphere incoming solar radiation,
atmospheric composition of those radiatively
active species that are not explicitly calculated in
the model (see Section 2.3.2.1), and distribution of
vegetation types (see Section 2.3.2.2). Most of
these quantities are relatively well observed for the
present climate state, and estimates exist for past
climates, allowing relatively easy interpretation of
model simulations of a particular climate state
(but see Section 2.3.4 below). However, for longtime scale climate variability and change there is a
potential for feedbacks of the biogeochemical
cycles on climate (Charlson et al., 1987; IPCC,
1995 (Chapter 9); Maier-Reimer et al., 1996;
Levis et al., 1999), and development of models
that include such feedbacks is still at an early stage
(Cox et al., 2000).
2.3 Coupled Ocean–Atmosphere Models
83
Wood and Bryan