interactions, must be included. While our focus
here is on the ocean, we include brief descriptions
of the representation of the other components in
coupled models. A more detailed discussion of the
fundamentals of climate models can be found in
Trenberth (1992), although the models have developed since this reference was written.
2.3.2.1 Atmosphere
Atmospheric GCMs are conceptually very similar
to oceanic GCMs (and of course are their historical forebears). They typically consist of a dynamical ‘core’ that time-steps the equations of motion
in discretized form, linked to a number of parameterizations of processes that are unresolved at the
scale of the discretization. One difference between
atmospheric and oceanic GCMs is that for the
atmosphere, the lack of side boundaries makes
spectral discretization methods attractive in the
horizontal, although finite difference methods are
also widely used (see the article by J. J. Hack in
Trenberth, 1992, for a fuller discussion of both
spectral and finite difference methods). Typical
resolution for a modern atmospheric model is T42
spectral truncation or about 3°. Since a typical
first baroclinic Rossby radius in the atmosphere is
around 700 km, this means that such models are
what in ocean modelling terms would be called
‘eddy permitting’; the models develop synoptic
scale mid-latitude weather systems, which are
known to play an important part in the general
circulation, but these systems cannot be said to be
fully resolved.
The vertical coordinate is usually pressure, height
or a terrain following (sigma) coordinate. Sometimes a generalized coordinate combining two or
more of these is used. Typically around 20–30
levels are used, unequally spaced to concentrate
attention in the regions of particular interest. Resolution is often enhanced in the surface boundary
layer, and many models have only a few levels
above the tropopause.
The sub-grid-scale processes which must be parameterized in atmospheric models (often referred to
as the ‘physics’ of the model) are perhaps more
numerous than in oceanic models (at least at the
present stage of model development). First, many
important dynamical processes are unresolved, yet
produce significant vertical transports of heat,
water vapour and momentum. Particular processes
that must be parameterized are convective plumes,
gravity waves and turbulent boundary layer eddies.
(See the chapter by J. T. Kiehl in Trenberth (1992)
for a more complete exposition of the fundamentals of parameterization of physical processes in
atmospheric climate GCMs.)
The water cycle is an important element of
atmospheric climate. Water exists in gaseous,
liquid and solid phases in the atmosphere, and its
transports and phase transformations (e.g. the formation of clouds and precipitation) must be accurately represented. Many of the key processes take
place at micrometre to millimetre scales and so
could not be resolved in the foreseeable future.
Clouds and water vapour have important radiative effects, absorbing, emitting, scattering and
reflecting the incoming solar and outgoing longwave radiation (see Bryden and Imawaki, Chapter 6.1). The representation of clouds and their
radiative properties is one of the largest areas of
uncertainty in climate modelling at present. Other
trace gases such as CO 2 and methane, and aerosols
(small airborne particles) from natural and anthropogenic sources, have important radiative effects,
which must be included in models. In current
coupled models, concentrations of most of these
trace substances are prescribed, although simple
representations of the physics and chemistry of
sulfate aerosols are now being developed (e.g.
Jones et al., 1999a).
Because of the greater complexity of subgrid-scale parameterizations in atmospheric models, compared with the current generation of ocean
models, atmospheric GCMs typically require considerably more computing resources per gridpoint
than oceanic GCMs. Yet many of the longer time
scales in the climate system are set by oceanic
processes. This means that for certain long-time
scale problems the use of ocean models coupled to
simplified or ‘intermediate complexity’ atmospheric models is attractive. These may be dynamical atmospheric GCMs that are highly truncated in
one or more directions (e.g. Saravanan et al., 2000),
or models in which the atmospheric dynamics and
associated transports of heat and water are completely parameterized or specified. Transports due
to baroclinic waves are known to be fundamental
to the atmospheric heat and water balances, whereas
the role of their oceanic counterparts (mesoscale
eddies) is still under debate (see, e.g. Wunsch,
1999b). These highly parameterized atmospheric
models vary in complexity from models with
SECTION 2 OBSERVATIONS AND MODELS
80
here is on the ocean, we include brief descriptions
of the representation of the other components in
coupled models. A more detailed discussion of the
fundamentals of climate models can be found in
Trenberth (1992), although the models have developed since this reference was written.
2.3.2.1 Atmosphere
Atmospheric GCMs are conceptually very similar
to oceanic GCMs (and of course are their historical forebears). They typically consist of a dynamical ‘core’ that time-steps the equations of motion
in discretized form, linked to a number of parameterizations of processes that are unresolved at the
scale of the discretization. One difference between
atmospheric and oceanic GCMs is that for the
atmosphere, the lack of side boundaries makes
spectral discretization methods attractive in the
horizontal, although finite difference methods are
also widely used (see the article by J. J. Hack in
Trenberth, 1992, for a fuller discussion of both
spectral and finite difference methods). Typical
resolution for a modern atmospheric model is T42
spectral truncation or about 3°. Since a typical
first baroclinic Rossby radius in the atmosphere is
around 700 km, this means that such models are
what in ocean modelling terms would be called
‘eddy permitting’; the models develop synoptic
scale mid-latitude weather systems, which are
known to play an important part in the general
circulation, but these systems cannot be said to be
fully resolved.
The vertical coordinate is usually pressure, height
or a terrain following (sigma) coordinate. Sometimes a generalized coordinate combining two or
more of these is used. Typically around 20–30
levels are used, unequally spaced to concentrate
attention in the regions of particular interest. Resolution is often enhanced in the surface boundary
layer, and many models have only a few levels
above the tropopause.
The sub-grid-scale processes which must be parameterized in atmospheric models (often referred to
as the ‘physics’ of the model) are perhaps more
numerous than in oceanic models (at least at the
present stage of model development). First, many
important dynamical processes are unresolved, yet
produce significant vertical transports of heat,
water vapour and momentum. Particular processes
that must be parameterized are convective plumes,
gravity waves and turbulent boundary layer eddies.
(See the chapter by J. T. Kiehl in Trenberth (1992)
for a more complete exposition of the fundamentals of parameterization of physical processes in
atmospheric climate GCMs.)
The water cycle is an important element of
atmospheric climate. Water exists in gaseous,
liquid and solid phases in the atmosphere, and its
transports and phase transformations (e.g. the formation of clouds and precipitation) must be accurately represented. Many of the key processes take
place at micrometre to millimetre scales and so
could not be resolved in the foreseeable future.
Clouds and water vapour have important radiative effects, absorbing, emitting, scattering and
reflecting the incoming solar and outgoing longwave radiation (see Bryden and Imawaki, Chapter 6.1). The representation of clouds and their
radiative properties is one of the largest areas of
uncertainty in climate modelling at present. Other
trace gases such as CO 2 and methane, and aerosols
(small airborne particles) from natural and anthropogenic sources, have important radiative effects,
which must be included in models. In current
coupled models, concentrations of most of these
trace substances are prescribed, although simple
representations of the physics and chemistry of
sulfate aerosols are now being developed (e.g.
Jones et al., 1999a).
Because of the greater complexity of subgrid-scale parameterizations in atmospheric models, compared with the current generation of ocean
models, atmospheric GCMs typically require considerably more computing resources per gridpoint
than oceanic GCMs. Yet many of the longer time
scales in the climate system are set by oceanic
processes. This means that for certain long-time
scale problems the use of ocean models coupled to
simplified or ‘intermediate complexity’ atmospheric models is attractive. These may be dynamical atmospheric GCMs that are highly truncated in
one or more directions (e.g. Saravanan et al., 2000),
or models in which the atmospheric dynamics and
associated transports of heat and water are completely parameterized or specified. Transports due
to baroclinic waves are known to be fundamental
to the atmospheric heat and water balances, whereas
the role of their oceanic counterparts (mesoscale
eddies) is still under debate (see, e.g. Wunsch,
1999b). These highly parameterized atmospheric
models vary in complexity from models with
SECTION 2 OBSERVATIONS AND MODELS
80