The oceanic general circulation models are also constructed based on fluid dynamics equations with the additional constraint of salt conservation. The discretization used
is in grid point because of the geometry of the basins. The
specificities of oceanic general circulation models are not
detailed further here.
Towards an ‘Integrated’ Model of the Earth
System
Historically, the first climate simulations carried out with
general circulation models employed ‘only’ atmospheric
models. Interactions with the surface, especially with the
ocean, were very limited because the majority of the surface
characteristics were imposed (surface temperature of the
oceans, presence of sea ice, surface albedo, roughness of
terrain etc.). As a result, the atmospheric circulation obtained
was in equilibrium with these surface conditions and other
forcings. In particular, it was then possible to evaluate the
response of the atmosphere to changes in the ocean surface.
It is clear, however, that the ocean, like vegetation and land
surfaces, does not remain unaffected when faced with climate change. Modelers therefore quickly sought to estimate
the impact of feedbacks from the other components of the
climate system on the atmosphere, which in turn defines the
climate at the surface. Figure 25.2 shows the evolution of
climate models since their inception. It shows the coupling
first with ocean surface models, then with complete ocean
circulation models. In parallel, land surface models have
progressed from simple hydrological models, with fixed
albedo and surface terrain, to models including interactive
vegetation, allowing the surface characteristics to be calculated according to changes in vegetation caused by changes
in climate or by man. Finally, models increasingly include a
representation of atmospheric chemistry and aerosols, which
have a significant influence on radiation, as well as the
biogeochemical cycles such as the carbon cycle. In this type
of model, the atmospheric concentration of CO 2 is no longer
imposed and is instead calculated from emissions.
Climate models developed in this way require enormous
computing power. Simulations are generally carried out on
supercomputers adapted to this type of coding. These computers are scarce, which explains the limited number of
general circulation models in the world. The models, their
resolution, as well as the components of the climate system
and the processes to be included are chosen at the outset
according to the issue to be addressed, but also in keeping
with the current limits in computing power. The performance of the models will vary depending on the model and
the computer used, but for example, the approximate computation times for the IPSL model are: run at very low resolution (IPSL-CM5A2: atmosphere 96 Â 95 Â 39, ocean
2°), about 70 years per day; at low resolution (IPSL-CM6LR, atmosphere 144 Â 133 Â 79, ocean 1°) 16 years per
day, and at medium resolution (IPSL-CM6-MR, atmosphere
280 Â 280 Â 79, ocean 1°) about 6 years per day.
Thanks to the improvements in supercomputers over
recent decades, the development of coupled atmosphereocean models, followed by atmosphere-ocean-vegetation
models has become possible. These models require a longer
computing time, not because there are many additional calculations to be performed for a given duration, but because
vegetation, and even more, the ocean, are components of the
climate system whose response time is far greater that of the
atmosphere. While we consider that simulations using an
atmospheric model alone, forced by boundary conditions
which repeat each year, must be integrated over a period of
20 to 50 years to obtain a response from the atmosphere
in equilibrium with these boundary conditions, a coupled
atmosphere-ocean model, in principle, needs to be integrated
over one or even several thousand years. The biggest challenge then is to close-off the water and energy balances in
the model to avoid a gradual drift related not to the imposed
forcing but to the model itself.
‘Realistic’ Modelling of Paleoclimates
Boundary Conditions and Initial Conditions
Many paleoclimate simulations aim to ‘recreate’ past climates as accurately as possible. The models can then be
evaluated in the context of climates documented by paleoclimate indicators which are different from the current climate. It also provides a better understanding of the possible
connections between differences in climate between distant
regions and supports reconstructions by providing better
spatial and temporal coverage or by including a regional
phenomenon not covered by the reconstructions, thereby
improving our understanding of them. We will return to the
comparison between models and data and the value of this
exercise at the end of the section.
How can a ‘realistic’ simulation of a paleoclimate be
achieved? First, depending on the question at stake, the part
of the climate system being assessed needs to be defined. For
example, if we want to study the continental climate in the
context of given ocean conditions, it is best to use an
atmospheric model, possibly coupled with a dynamic land
surface model or vegetation model. Once the subsystem is
selected, a paleoclimate simulation is carried out by
imposing the most realistic forcings and boundary conditions possible on this subsystem for the simulation period.
Thus, the more the subsystem is constrained, the more
conditions there are to be imposed, conditions that need to
be known for the study period. Continuing with the example
of the simulation using only a general atmospheric
328
M. Kageyama and D. Paillard
is in grid point because of the geometry of the basins. The
specificities of oceanic general circulation models are not
detailed further here.
Towards an ‘Integrated’ Model of the Earth
System
Historically, the first climate simulations carried out with
general circulation models employed ‘only’ atmospheric
models. Interactions with the surface, especially with the
ocean, were very limited because the majority of the surface
characteristics were imposed (surface temperature of the
oceans, presence of sea ice, surface albedo, roughness of
terrain etc.). As a result, the atmospheric circulation obtained
was in equilibrium with these surface conditions and other
forcings. In particular, it was then possible to evaluate the
response of the atmosphere to changes in the ocean surface.
It is clear, however, that the ocean, like vegetation and land
surfaces, does not remain unaffected when faced with climate change. Modelers therefore quickly sought to estimate
the impact of feedbacks from the other components of the
climate system on the atmosphere, which in turn defines the
climate at the surface. Figure 25.2 shows the evolution of
climate models since their inception. It shows the coupling
first with ocean surface models, then with complete ocean
circulation models. In parallel, land surface models have
progressed from simple hydrological models, with fixed
albedo and surface terrain, to models including interactive
vegetation, allowing the surface characteristics to be calculated according to changes in vegetation caused by changes
in climate or by man. Finally, models increasingly include a
representation of atmospheric chemistry and aerosols, which
have a significant influence on radiation, as well as the
biogeochemical cycles such as the carbon cycle. In this type
of model, the atmospheric concentration of CO 2 is no longer
imposed and is instead calculated from emissions.
Climate models developed in this way require enormous
computing power. Simulations are generally carried out on
supercomputers adapted to this type of coding. These computers are scarce, which explains the limited number of
general circulation models in the world. The models, their
resolution, as well as the components of the climate system
and the processes to be included are chosen at the outset
according to the issue to be addressed, but also in keeping
with the current limits in computing power. The performance of the models will vary depending on the model and
the computer used, but for example, the approximate computation times for the IPSL model are: run at very low resolution (IPSL-CM5A2: atmosphere 96 Â 95 Â 39, ocean
2°), about 70 years per day; at low resolution (IPSL-CM6LR, atmosphere 144 Â 133 Â 79, ocean 1°) 16 years per
day, and at medium resolution (IPSL-CM6-MR, atmosphere
280 Â 280 Â 79, ocean 1°) about 6 years per day.
Thanks to the improvements in supercomputers over
recent decades, the development of coupled atmosphereocean models, followed by atmosphere-ocean-vegetation
models has become possible. These models require a longer
computing time, not because there are many additional calculations to be performed for a given duration, but because
vegetation, and even more, the ocean, are components of the
climate system whose response time is far greater that of the
atmosphere. While we consider that simulations using an
atmospheric model alone, forced by boundary conditions
which repeat each year, must be integrated over a period of
20 to 50 years to obtain a response from the atmosphere
in equilibrium with these boundary conditions, a coupled
atmosphere-ocean model, in principle, needs to be integrated
over one or even several thousand years. The biggest challenge then is to close-off the water and energy balances in
the model to avoid a gradual drift related not to the imposed
forcing but to the model itself.
‘Realistic’ Modelling of Paleoclimates
Boundary Conditions and Initial Conditions
Many paleoclimate simulations aim to ‘recreate’ past climates as accurately as possible. The models can then be
evaluated in the context of climates documented by paleoclimate indicators which are different from the current climate. It also provides a better understanding of the possible
connections between differences in climate between distant
regions and supports reconstructions by providing better
spatial and temporal coverage or by including a regional
phenomenon not covered by the reconstructions, thereby
improving our understanding of them. We will return to the
comparison between models and data and the value of this
exercise at the end of the section.
How can a ‘realistic’ simulation of a paleoclimate be
achieved? First, depending on the question at stake, the part
of the climate system being assessed needs to be defined. For
example, if we want to study the continental climate in the
context of given ocean conditions, it is best to use an
atmospheric model, possibly coupled with a dynamic land
surface model or vegetation model. Once the subsystem is
selected, a paleoclimate simulation is carried out by
imposing the most realistic forcings and boundary conditions possible on this subsystem for the simulation period.
Thus, the more the subsystem is constrained, the more
conditions there are to be imposed, conditions that need to
be known for the study period. Continuing with the example
of the simulation using only a general atmospheric
328
M. Kageyama and D. Paillard