Long-term Climate Projections - Climate Change: Identification and Projections - page 112

factors, including the carbon cycle, and also on the absence of greenhouse
effect saturation.

Should we use the term projections or extrapolations? In the case of
CMIP5 the simulated models owe nothing to previous observations: future
simulations owe almost everything to the scenarios selected and the pre-
determined knowledge-based models. The term projection would therefore
seem more appropriate.

In our case, it is legitimate to use the term extrapolation. The future
remains dependent on the scenario of CO 2 concentration, but the model itself
owes everything to the paleoclimatic observations. By means of comparison,
when we carry out a linear extrapolation (

)( ), coefficients a
and b constitute, to some extent, a model, identified by linear regression of
the recent past of the observed signal. In our case, the model (EBM) is more
complex and features input from exogenous signals, but its coefficients are
determined using the signal to be extrapolated, and are only involved as
intermediary results in the calculation. The only a priori knowledge involved
is the energy balance structure (no more arbitrary than the affine structure of
the linear extrapolations), and the minor constraint

y t + τ = a τ + b

S

> 0

, which is

lim

usually inactive.

The simulations in graph (b) also constitute extrapolations according to
our reasoning, but they use the strict constraint

− 2

S irr , which
arises from an a priori certainty, without which the observations would
contradict the alarmist assumptions of the IPCC.

< 1 62 ° C / Wm

The ranges in the two graphs in Figure 10.3 must not be interpreted as
being representative of the calculated confidence intervals. They simply
reflect the discordance of the various paleoclimatic reconstructions from
which they are derived. The ranges indicated for the IPCC projections
(Figure 10.1) are more formal (+/- 1.64 standard deviation). However, they
are only as reliable as the prior uncertainties on the models' climatic
reactions and, even earlier, on the radiative forcing incorporated in the RCPs
(AR5, Figure 12.4).