Short-term Predictions - Climate Change: Identification and Projections - page 119

- the effects of model errors applied to inputs (structural and parametric
errors).
A model having been established ( a priori or identified) as well as the
input scenarios, a long-term prediction is reduced to a simple simulation,
while a short-term prediction requires an extrapolation of the output error to
be added to the simulation, whether this is totally due to internal climate
variability or not.
Consequently, let us revisit such an output error (Figure 11.2), extracted
from Figure 7.6.
Figure 11.2. The output error: image of random climate fluctuations
The above line is zero mean, but it shows some continuity: it is not white
noise. It shows no regular periodicity, apart from perhaps over the last two-
hundred years, but such periodicity is dubious given that it was either absent
or hidden for the previous 1000 years. It cannot be confirmed elsewhere on
the correlation function (see Figure 7.10). At first glance, when the signal is
clearly positive, we would expect it to return to the average value in 10 or 20
years. We are not able to go any further than this, and are left with a null
prediction.
According to Figure 11.2, it appears that around the year 2000, the
natural fluctuation was clearly positive. We could therefore expect a drop in
this level in the coming years. The theory of the IPCC is precisely the
opposite, maintaining that the current stagnation is due to the (temporary)
cancelling out of a strong, deterministic warming effect (caused by CO 2 ) by
a natural negative fluctuation. This point is contradicted, not only by
Figure 11.2, but by all of the historic observations, processed together by all
the models, free or forced. Indeed, let us consider Figure 11.3, which