selection of this learning period (Jansen 2007) and the
number of statistical components to be used, but the result in
the end is not very sensitive to this selection. During this
period, a linear regression is performed between the proxies
and the statistical characteristics observed. This regression
attaches weights to proxies in order to maximize their correlation with temperatures. This analysis also makes it possible to eliminate the proxies which have a poor correlation
with the temperature signal during the learning period.
A ‘verification’ period prior to the learning period for
which instrumental temperature data are available is then
used. This verification period identifies the errors incurred in
the regression obtained for the calibration period. For this
verification period, it is also possible to determine the error
linked to the omission of proxies in the estimation of the
temperature. This calculation is essential because it is
obvious that the further back in time we go, the fewer climate series are available and the more uncertain the reconstruction of temperatures becomes.
Finally, temperatures can be reconstructed for the last
millennium. It should be noted that this reconstruction has a
spatial aspect. It is important to bear in mind that this type of
method relies on basic assumptions about the temporal stability of the climate modes identified in the temperature data
during the calibration period and the temporal stability of the
relationship between proxies and temperatures. These two
types of stability can be quantified over the verification
period, but it is impossible to exclude the possibility of
changes over a longer period. Another problem with this
type of reconstruction comes from the statistical regression
between proxies and main components over the calibration
period. Since a regression is generally imperfect, it inevitably leads to an underestimation of the variance when this is
used to reconstruct the climate, which can lead to a poor
estimation of long-term climate variations. It is possible to
partially solve this problem by using proxies that represent
different time scales, and are thus sensitive to scales varying
from inter-annual to centennial (Jansen 2007).
A major advance in the quality of climate reconstructions
lies in the improved understanding of the mechanism linking
the ‘proxy’ to the climate variation, which helps the statistical steps described above to be guided by knowledge of the
physics. This research topic is particularly active at the
moment.
Several research teams have proposed temperature
reconstructions for the last millennium, based on different
proxy datasets (Jansen 2007). These reconstructions often
have common foundations (often, tree ring data) but the
spatial distribution of the proxies used varies considerably
from one reconstruction to another.
The evolution of the error bars of these reconstructions
shown in Fig. 30.4 shows the discrepancies between the
estimates of temperature changes, especially during cold
periods. It should be noted in particular that the latter part of
the twentieth century emerges significantly from the error bars
of the temperature variations for the preceding millennium.
Climate Simulations
Climate simulations provide an understanding of how different forcings affect the climate and quantify the main
feedbacks. Moreover, comparison between the model results
and the data makes it possible to determine whether the
models are capable of representing a climate different from
the current one. The models used for Holocene simulations
cover the spectrum of models presented in Chap. 4 of this
volume which describes the different hypotheses and the
protocols to run such simulations. In the case of the Holocene, it is mainly the characteristics of the seasonal cycle that
have been analyzed, as changes in insolation, driven by
precession, strongly modulate seasonality but have little
impact on the annual average of the different climate variables. For the last 2000 years, the emphasis is on understanding the forcing associated with fluctuations in the solar
constant and in volcanism, and the identification of associated feedbacks. This is a major step towards gaining perspective on recent centuries strongly disrupted by human
activity. Increased understanding of interannual to centennial
variability is necessary in order to detect climate change and,
where appropriate, to attribute it to human activities.
Holocene Simulations
Major Trends
There are very few simulations that cover the entire Holocene. Those that exist were carried out using models of
intermediate complexity (see Chap 4), because it is impossible to simulate a period of 10,000 years within an
acceptable time frame using general circulation models,
which need an average of one month to achieve 100 years of
simulation on supercomputers.
The main objective of simulations of the whole Holocene
period is to reproduce the major climate trends caused by
variations in orbital parameters and by the concentration of
greenhouse gases. They generally do not take into consideration the full set of forcings such as volcanism or the
evolution of the solar constant, factors which are not well
known for the whole period. The applied models of intermediate complexity take account of atmospheric and ocean
circulation in a simplified way, as well as sea ice and vegetation (Crucifix et al. 2002; Renssen et al. 2005). They
nevertheless have different levels of complexity.
The first model considers the ocean and the atmosphere in
terms of latitudinal sectors, while the second takes a
430
P. Braconnot and P. Yiou
number of statistical components to be used, but the result in
the end is not very sensitive to this selection. During this
period, a linear regression is performed between the proxies
and the statistical characteristics observed. This regression
attaches weights to proxies in order to maximize their correlation with temperatures. This analysis also makes it possible to eliminate the proxies which have a poor correlation
with the temperature signal during the learning period.
A ‘verification’ period prior to the learning period for
which instrumental temperature data are available is then
used. This verification period identifies the errors incurred in
the regression obtained for the calibration period. For this
verification period, it is also possible to determine the error
linked to the omission of proxies in the estimation of the
temperature. This calculation is essential because it is
obvious that the further back in time we go, the fewer climate series are available and the more uncertain the reconstruction of temperatures becomes.
Finally, temperatures can be reconstructed for the last
millennium. It should be noted that this reconstruction has a
spatial aspect. It is important to bear in mind that this type of
method relies on basic assumptions about the temporal stability of the climate modes identified in the temperature data
during the calibration period and the temporal stability of the
relationship between proxies and temperatures. These two
types of stability can be quantified over the verification
period, but it is impossible to exclude the possibility of
changes over a longer period. Another problem with this
type of reconstruction comes from the statistical regression
between proxies and main components over the calibration
period. Since a regression is generally imperfect, it inevitably leads to an underestimation of the variance when this is
used to reconstruct the climate, which can lead to a poor
estimation of long-term climate variations. It is possible to
partially solve this problem by using proxies that represent
different time scales, and are thus sensitive to scales varying
from inter-annual to centennial (Jansen 2007).
A major advance in the quality of climate reconstructions
lies in the improved understanding of the mechanism linking
the ‘proxy’ to the climate variation, which helps the statistical steps described above to be guided by knowledge of the
physics. This research topic is particularly active at the
moment.
Several research teams have proposed temperature
reconstructions for the last millennium, based on different
proxy datasets (Jansen 2007). These reconstructions often
have common foundations (often, tree ring data) but the
spatial distribution of the proxies used varies considerably
from one reconstruction to another.
The evolution of the error bars of these reconstructions
shown in Fig. 30.4 shows the discrepancies between the
estimates of temperature changes, especially during cold
periods. It should be noted in particular that the latter part of
the twentieth century emerges significantly from the error bars
of the temperature variations for the preceding millennium.
Climate Simulations
Climate simulations provide an understanding of how different forcings affect the climate and quantify the main
feedbacks. Moreover, comparison between the model results
and the data makes it possible to determine whether the
models are capable of representing a climate different from
the current one. The models used for Holocene simulations
cover the spectrum of models presented in Chap. 4 of this
volume which describes the different hypotheses and the
protocols to run such simulations. In the case of the Holocene, it is mainly the characteristics of the seasonal cycle that
have been analyzed, as changes in insolation, driven by
precession, strongly modulate seasonality but have little
impact on the annual average of the different climate variables. For the last 2000 years, the emphasis is on understanding the forcing associated with fluctuations in the solar
constant and in volcanism, and the identification of associated feedbacks. This is a major step towards gaining perspective on recent centuries strongly disrupted by human
activity. Increased understanding of interannual to centennial
variability is necessary in order to detect climate change and,
where appropriate, to attribute it to human activities.
Holocene Simulations
Major Trends
There are very few simulations that cover the entire Holocene. Those that exist were carried out using models of
intermediate complexity (see Chap 4), because it is impossible to simulate a period of 10,000 years within an
acceptable time frame using general circulation models,
which need an average of one month to achieve 100 years of
simulation on supercomputers.
The main objective of simulations of the whole Holocene
period is to reproduce the major climate trends caused by
variations in orbital parameters and by the concentration of
greenhouse gases. They generally do not take into consideration the full set of forcings such as volcanism or the
evolution of the solar constant, factors which are not well
known for the whole period. The applied models of intermediate complexity take account of atmospheric and ocean
circulation in a simplified way, as well as sea ice and vegetation (Crucifix et al. 2002; Renssen et al. 2005). They
nevertheless have different levels of complexity.
The first model considers the ocean and the atmosphere in
terms of latitudinal sectors, while the second takes a
430
P. Braconnot and P. Yiou