Overall Results - Climate Change: Identification and Projections - page 92

Moreover, it is traditional practice to use different series of
measurements, occasionally of various qualities. As a general rule, they are
implicitly assumed to be valid: not affected by bias or significant calibration
errors. If such defects are present, uncertainty calculations on identified
parameters no longer hold, due to a lack of ergodicity (condition mentioned
at the end of paragraph 6.3). Indeed, the major difficulty encountered in the
paleoclimatic reconstructions regards the calibration of the proxies used.
When the calibration is incorrect, due to the choice of proxies or the method
of statistical processing, an error is introduced which cannot be compensated
for by average over time, carried out through the identifications.
Prior to the technical hypothesis of ergodicity, it goes without saying that
the identification data is supposed to reflect the same process. In this case,
this would occasionally seem rather doubtful. The hockey stick curve, which
ignores large climatic events, seems to have come straight from another
world; one of a pre-industrial lost paradise often mentioned by Sir John
Houghton (Figure 3.1). Similarly, the ER background corrections are
explicitly inspired by extrasolar systems.
Finally, if the reconstructions lead to disconnected regions of uncertainty,
we can only take note of this and reject all or part of the conclusions derived,
either from one or from the others. This selective rejection must therefore be
based on criteria other than those of identification, for instance the level of
confidence ( low, medium, high ) given to one type of reconstruction with
regard to another. But who is empowered to make such distinctions?
Moreover, the uncertainties calculated are only approximations,
asymptotically valid when the identification durations tends to infinity,
which is relatively far from being achieved. They also assume the time
invariance of statistical characteristics of noises and measurement errors,
which are significantly different, depending on whether modern
measurements or paloeclimatic measurements are being used. In the end,
even if the thresholds and confidence intervals only have approximate
values, we accept that they constitute more objective assessment criteria than
the assertion of confidence from experts, assured only by their own personal
judgment.