Partial Results - Climate Change: Identification and Projections - page 87

At the top, the functions of autocovariance have been traced. The initial
dot shows the value of the variances. It is necessarily greater in the case of
forced identification (right), where the minimization lacks the degree of
freedom S 2 . The question is to know whether this increase is significant. In
both cases, the error is far from being a white noise (for which the
autocorrelation function would be an impulse), but the “coloring” implied by
the wider range is more pronounced for the forced identification. And above
all, the presence of a residual cross-correlation with the irradiance input (in a
dotted-line box) shows the causality which the constrained model does not
sufficiently account for.

Estimated confidence intervals

For all of the following, the confidence intervals and regions associated
to the identified parameters are given at 90% ( very likely, according to the
IPCC terminology). This corresponds to confidence intervals at

± 1 . 645

standard deviation.

For climate sensitivity, we have, in accordance with the calculation
methods in section 6.2:

(90%) S = 1.28 [ − 0.53 à 3.09] ° C

c lim

c S . It confirms that the
hypothesis of zero climate sensitivity is not incompatible with the
combination of paleoclimatic data used.

The confidence interval covers the value = 0

lim

The upper value of the confidence range obtained is much lower than the
extreme limit (6°C) stated by the IPCC. This value is unrealistic given the
data processed.

These results are already significant, but the range is too wide for the
estimated value to constitute a reliable indicator.

For sensitivity to net irradiance S 2 (also noted as S irr ), we have:

(90%)

W m −

2

S = 17.5 [10.6 à 24.4] ° C /

irr

This time, we have a robust indication: the low sensitivity to solar
activity stated by the IPCC ( ) is strictly incompatible S < 1.62 C / ° W m −

2

irr