Reconstructing and Modelling the Last Glacial Maximum
691
boundary conditions, the complete a1mospheric state
can be computed, thus yielding global fields
of quantities on which no or only sparse data are
available, e.g. temperature, precipitation, wind,
cloud cover etc. Lorenz et al. (1996) present results for the time slices 6, 21 and 115 ka BP
obtained with the ECHAM 3 model (DKRZ
Modellbetreuungsgruppe 1994). Especially quantities for ocean model forcing - wind stress and
surface heat and fresh water flux - can be obtained
by a1mospheric modelling. Within CLIMAP 2000,
a special benefit of a1mospheric models is their ability to connect the marine data on SSTs, which are
needed as a boundary condition, with terrestrial data
on near surface temperature and precipitation,
which can be used for validation. To assess the errors of a1mospheric modelling, the reproduction of
the modem climate by different models has been
compared within the Atmospheric Model
Intercomparison Project (AMIP, see Gates 1992).
Because all models have to some extent been
"tuned" to the modem climate, the results of AMIP
are not completely transferable to simulations of
other climate states, e.g. increased CO 2 scenarios
of representing anthropogenic greenhouse warming. To compare the reproduction of climate states
deviating from the modem one, several a1mospheric
models from different countries participate in the
Paleoclimate Modeling Intercomparison Project
(PMIP, Joussaume and Taylor 1995). Within this
project, the time slices 6 and 21 ka BP (LGM) were
simulated based on boundary conditions which are
the same for all models. For the LGM, lower boundary conditions were taken from the CLIMAP reconstruction. However, the vegetation has not been
changed.
As an example of the results, annual mean surface air temperature for the modem climate and
the LGM, computed by averaging six ofthe models participating in PMlP, are presented in Figs. 2
and 3 (cf. Wyputta and Grieger 1997). The largest
differences between the glacial and the modem
model runs of about -24°C occur over the European and North American ice-sheets. Also shown
are the local standard deviations between the model
results, which represent model uncertainties. The
largest differences among the six models occur
over the European and North American ice-sheets
and in the Antarctic region.
While over the oceans the temperatures are
closely tied to the prescribed SSTs, the global mean
standard deviation between the models over land
amounts to 4.0°C for the control runs and to 5.1 °C
for the LGM runs. The larger latter value indicates
that model uncertainties are indeed increased for
climate states different from the modem one. The
root-mean-square temperature difference over land
between LGM and control runs is 7.8°C, implying
a "signal to noise ratio" larger than unity in the model
response.
Ocean Modelling
Equilibrium ocean-alone solutions for the glacial
circulation all rest on the CLIMAP data set: They
require one or more products derived from standalone runs of a1mospheric models which in tum
have used the surface boundary conditions set by
CLIMAP. Moreover, the majority of these solutions
use a restoring boundary condition in which the
ocean model upper layer temperature is continuously restored to the CLIMAP-SST estimates.
All ocean models are dynamically forced by
surface wind stress from a1mospheric models. Bigg
(1994) and Winguth et al. (1996) chose to calculate the surface heat flux from the glacial response
of atmospheric models as well. In contrast,
Lautenschlager et al. (1992) and Fichefet et al.
(1994) decided for directly restoring to the
CLIMAP-SST data. Haupt et al. (1994, 1995),
Seidov and Prien (1996) and Seidov et al. (1996)
did so, too, but modified the CLIMAP-SST data in
the northern North Atlantic according to Samthein
et al. (1994,1995).
Unfortunately there is no global set of sea surface salinity (SSS) estimates for the LGM to which
the model upper layer salinity could be restored to.
This problem can be dealt with either by using the
fresh water flux from atmospheric models
(Lautenschlageretal. 1992; Bigg 1994) or by generating some composite SSS data (Fichefet et al.
1994; Seidov and Prien 1996; Seidov et al.
1 996;Winguth et al. 1996). These composite SSS
data consist of modem Levitus (1982) or Levitus
691
boundary conditions, the complete a1mospheric state
can be computed, thus yielding global fields
of quantities on which no or only sparse data are
available, e.g. temperature, precipitation, wind,
cloud cover etc. Lorenz et al. (1996) present results for the time slices 6, 21 and 115 ka BP
obtained with the ECHAM 3 model (DKRZ
Modellbetreuungsgruppe 1994). Especially quantities for ocean model forcing - wind stress and
surface heat and fresh water flux - can be obtained
by a1mospheric modelling. Within CLIMAP 2000,
a special benefit of a1mospheric models is their ability to connect the marine data on SSTs, which are
needed as a boundary condition, with terrestrial data
on near surface temperature and precipitation,
which can be used for validation. To assess the errors of a1mospheric modelling, the reproduction of
the modem climate by different models has been
compared within the Atmospheric Model
Intercomparison Project (AMIP, see Gates 1992).
Because all models have to some extent been
"tuned" to the modem climate, the results of AMIP
are not completely transferable to simulations of
other climate states, e.g. increased CO 2 scenarios
of representing anthropogenic greenhouse warming. To compare the reproduction of climate states
deviating from the modem one, several a1mospheric
models from different countries participate in the
Paleoclimate Modeling Intercomparison Project
(PMIP, Joussaume and Taylor 1995). Within this
project, the time slices 6 and 21 ka BP (LGM) were
simulated based on boundary conditions which are
the same for all models. For the LGM, lower boundary conditions were taken from the CLIMAP reconstruction. However, the vegetation has not been
changed.
As an example of the results, annual mean surface air temperature for the modem climate and
the LGM, computed by averaging six ofthe models participating in PMlP, are presented in Figs. 2
and 3 (cf. Wyputta and Grieger 1997). The largest
differences between the glacial and the modem
model runs of about -24°C occur over the European and North American ice-sheets. Also shown
are the local standard deviations between the model
results, which represent model uncertainties. The
largest differences among the six models occur
over the European and North American ice-sheets
and in the Antarctic region.
While over the oceans the temperatures are
closely tied to the prescribed SSTs, the global mean
standard deviation between the models over land
amounts to 4.0°C for the control runs and to 5.1 °C
for the LGM runs. The larger latter value indicates
that model uncertainties are indeed increased for
climate states different from the modem one. The
root-mean-square temperature difference over land
between LGM and control runs is 7.8°C, implying
a "signal to noise ratio" larger than unity in the model
response.
Ocean Modelling
Equilibrium ocean-alone solutions for the glacial
circulation all rest on the CLIMAP data set: They
require one or more products derived from standalone runs of a1mospheric models which in tum
have used the surface boundary conditions set by
CLIMAP. Moreover, the majority of these solutions
use a restoring boundary condition in which the
ocean model upper layer temperature is continuously restored to the CLIMAP-SST estimates.
All ocean models are dynamically forced by
surface wind stress from a1mospheric models. Bigg
(1994) and Winguth et al. (1996) chose to calculate the surface heat flux from the glacial response
of atmospheric models as well. In contrast,
Lautenschlager et al. (1992) and Fichefet et al.
(1994) decided for directly restoring to the
CLIMAP-SST data. Haupt et al. (1994, 1995),
Seidov and Prien (1996) and Seidov et al. (1996)
did so, too, but modified the CLIMAP-SST data in
the northern North Atlantic according to Samthein
et al. (1994,1995).
Unfortunately there is no global set of sea surface salinity (SSS) estimates for the LGM to which
the model upper layer salinity could be restored to.
This problem can be dealt with either by using the
fresh water flux from atmospheric models
(Lautenschlageretal. 1992; Bigg 1994) or by generating some composite SSS data (Fichefet et al.
1994; Seidov and Prien 1996; Seidov et al.
1 996;Winguth et al. 1996). These composite SSS
data consist of modem Levitus (1982) or Levitus