account the main forcings for the climate of the Last Glacial
Maximum: the expanse and altitude of the ice sheets, ocean
surface conditions and atmospheric concentration of CO 2 . It
was a simulation derived from one of the most sophisticated
stand-alone atmospheric models of that time. The simulated
duration was short (three months for the first simulation!)
compared with current norms: theoretically, a few decades of
simulations are required to obtain statistically robust results,
depending on the region and the phenomenon in question. It
must be recognized that at the time of the first paleoclimate
simulations, the duration of the simulations carried out with
general circulation models was greatly restricted by the cost
of computing time. To save time, experiments were carried
out under unchanging January or July sunshine conditions,
producing significant results for relatively short durations of
simulations (typically 90 days). These results, which were a
technical feat at the time of their publication, would now
probably be viewed very critically, mainly due to their short
duration, even in conditions of fixed insolation. Models are
evolving as our understanding of the climate system and
computer capabilities improves. Simulations therefore need
to be revised periodically in the light of these advances. The
simulations of the first phase of the PMIP project (Paleoclimate Modeling Intercomparison Project, Joussaume and
Taylor 1995) using general circulation models, ran for at
least ten years after an adjustment for the boundary conditions of at least one year.
These first simulations of the climate of the Last Glacial
Maximum used the ocean surface conditions reconstructed
by the CLIMAP project (1976, 1981). These reconstructions
were the result of a major work of data synthesis, but
problems were quickly identified, particularly for the subtropical regions, where a higher temperature than is the case
currently was reconstructed, and for the North Atlantic,
where the winter sea ice cover was overestimated. These
problems were partially amplified by the methods used to
extrapolate the reconstructions for each site to cover the
globe with an even grid. The CLIMAP project, at the
specific request of modelers, provided reconstructions for the
months of February and August. However, it is entirely
possible that certain species used for reconstructing SSTs
are not particularly sensitive to these specific months, but to
other factors. Thus, manipulating data to construct boundary
conditions for models, especially atmospheric ones, can
prove to be extremely restrictive for the interpretation of data
records. Furthermore, since they are used to establish the
boundary conditions, they cannot also be used to validate the
model. Therefore, as soon as they became available, it was
very useful to use coupled ocean-atmosphere models to
simulate paleoclimates. It is worth highlighting again the
challenge represented by the first coupled simulations of the
climate of the LGM. Again, in this case, the first published
simulation was only about thirty years long, a very short
time frame compared to the response time of the deep ocean!
Within the international PMIP2 project, eight groups have
carried out multi-centennial coupled ocean-atmosphere
simulations, which shows how difficult this type of experiment is to perform. This was confirmed in the 3rd phase of
the PMIP, contemporary with CMIP5, for which 9 models
have finally yielded results for the LGM. For PMIP4, there
are about fifteen modeling groups planning to undertake this
simulation, the future will tell us how many succeed.
This example of modeling of the LGM climate shows that
‘realistic’ modeling of this climate has evolved in line with
the forcings and tools available.
We have seen that the uncertainties in a ‘realistic’ simulation of a climate are due to two types of factors: those
related to the formulation of the model and those inherent in
the selection of boundary conditions. The next two sections
show how to quantify these uncertainties, both through
rigorous comparisons between the results of models forced
by identical boundary conditions, and through sensitivity
studies with respect to these boundary conditions.
Comparing results from different models:
Modeling Intercomparison Projects
How can the results of different models be compared? These
differences may be due to the models themselves, or to the
boundary conditions and forcings imposed on these models.
The results of several models can only be rigorously compared by assigning them the same boundary conditions/
forcings. Such exercises have been proposed for the modeling
of current climates using atmospheric general circulation
models (AMIP project, Atmospheric Model Intercomparison
Project, http://www-pcmdi.llnl.gov/projects/amip/), followed
by coupled models, both for current and future climates
(CMIP project, Coupled Model Intercomparison Project,
http://www-pcmdi.llnl.gov/projects/cmip/). The CMIP5 exercise corresponds to the results produced for the 5th IPCC
report and CMIP6, currently underway, will provide its first
results for the 6th IPCC report, which will be published in
2022. In the same vein, PMIP (Paleoclimate Modeling
Intercomparison Project), the project to compare paleoclimate
models came into being in the 1990s (http://pmip.lsce.ipsl.fr).
At first, this project involved atmospheric general circulation
models (PMIP1 project, http://pmip1.lsce.ipsl.fr/) for the
Middle Holocene (6000 years ago) and the Last Glacial
Maximum (21,000 years ago). It was then extended to coupled atmosphere-ocean and atmosphere-ocean-vegetation
models (PMIP2 project, Braconnot et al. (2007a, b), http://
pmip2.lsce.ipsl.fr/). A new feature of PMIP3 was to use climate models strictly identical to those used for CMIP5.
PMIP4 coordinates both CMIP6 simulations, which will
therefore use the same models as those used for climate
projections, and simulations based on other models, usually
longer ones or for older climates. The PMIP4-CMIP6
330
M. Kageyama and D. Paillard
Maximum: the expanse and altitude of the ice sheets, ocean
surface conditions and atmospheric concentration of CO 2 . It
was a simulation derived from one of the most sophisticated
stand-alone atmospheric models of that time. The simulated
duration was short (three months for the first simulation!)
compared with current norms: theoretically, a few decades of
simulations are required to obtain statistically robust results,
depending on the region and the phenomenon in question. It
must be recognized that at the time of the first paleoclimate
simulations, the duration of the simulations carried out with
general circulation models was greatly restricted by the cost
of computing time. To save time, experiments were carried
out under unchanging January or July sunshine conditions,
producing significant results for relatively short durations of
simulations (typically 90 days). These results, which were a
technical feat at the time of their publication, would now
probably be viewed very critically, mainly due to their short
duration, even in conditions of fixed insolation. Models are
evolving as our understanding of the climate system and
computer capabilities improves. Simulations therefore need
to be revised periodically in the light of these advances. The
simulations of the first phase of the PMIP project (Paleoclimate Modeling Intercomparison Project, Joussaume and
Taylor 1995) using general circulation models, ran for at
least ten years after an adjustment for the boundary conditions of at least one year.
These first simulations of the climate of the Last Glacial
Maximum used the ocean surface conditions reconstructed
by the CLIMAP project (1976, 1981). These reconstructions
were the result of a major work of data synthesis, but
problems were quickly identified, particularly for the subtropical regions, where a higher temperature than is the case
currently was reconstructed, and for the North Atlantic,
where the winter sea ice cover was overestimated. These
problems were partially amplified by the methods used to
extrapolate the reconstructions for each site to cover the
globe with an even grid. The CLIMAP project, at the
specific request of modelers, provided reconstructions for the
months of February and August. However, it is entirely
possible that certain species used for reconstructing SSTs
are not particularly sensitive to these specific months, but to
other factors. Thus, manipulating data to construct boundary
conditions for models, especially atmospheric ones, can
prove to be extremely restrictive for the interpretation of data
records. Furthermore, since they are used to establish the
boundary conditions, they cannot also be used to validate the
model. Therefore, as soon as they became available, it was
very useful to use coupled ocean-atmosphere models to
simulate paleoclimates. It is worth highlighting again the
challenge represented by the first coupled simulations of the
climate of the LGM. Again, in this case, the first published
simulation was only about thirty years long, a very short
time frame compared to the response time of the deep ocean!
Within the international PMIP2 project, eight groups have
carried out multi-centennial coupled ocean-atmosphere
simulations, which shows how difficult this type of experiment is to perform. This was confirmed in the 3rd phase of
the PMIP, contemporary with CMIP5, for which 9 models
have finally yielded results for the LGM. For PMIP4, there
are about fifteen modeling groups planning to undertake this
simulation, the future will tell us how many succeed.
This example of modeling of the LGM climate shows that
‘realistic’ modeling of this climate has evolved in line with
the forcings and tools available.
We have seen that the uncertainties in a ‘realistic’ simulation of a climate are due to two types of factors: those
related to the formulation of the model and those inherent in
the selection of boundary conditions. The next two sections
show how to quantify these uncertainties, both through
rigorous comparisons between the results of models forced
by identical boundary conditions, and through sensitivity
studies with respect to these boundary conditions.
Comparing results from different models:
Modeling Intercomparison Projects
How can the results of different models be compared? These
differences may be due to the models themselves, or to the
boundary conditions and forcings imposed on these models.
The results of several models can only be rigorously compared by assigning them the same boundary conditions/
forcings. Such exercises have been proposed for the modeling
of current climates using atmospheric general circulation
models (AMIP project, Atmospheric Model Intercomparison
Project, http://www-pcmdi.llnl.gov/projects/amip/), followed
by coupled models, both for current and future climates
(CMIP project, Coupled Model Intercomparison Project,
http://www-pcmdi.llnl.gov/projects/cmip/). The CMIP5 exercise corresponds to the results produced for the 5th IPCC
report and CMIP6, currently underway, will provide its first
results for the 6th IPCC report, which will be published in
2022. In the same vein, PMIP (Paleoclimate Modeling
Intercomparison Project), the project to compare paleoclimate
models came into being in the 1990s (http://pmip.lsce.ipsl.fr).
At first, this project involved atmospheric general circulation
models (PMIP1 project, http://pmip1.lsce.ipsl.fr/) for the
Middle Holocene (6000 years ago) and the Last Glacial
Maximum (21,000 years ago). It was then extended to coupled atmosphere-ocean and atmosphere-ocean-vegetation
models (PMIP2 project, Braconnot et al. (2007a, b), http://
pmip2.lsce.ipsl.fr/). A new feature of PMIP3 was to use climate models strictly identical to those used for CMIP5.
PMIP4 coordinates both CMIP6 simulations, which will
therefore use the same models as those used for climate
projections, and simulations based on other models, usually
longer ones or for older climates. The PMIP4-CMIP6
330
M. Kageyama and D. Paillard