From our results, three main
conclusions may be emphasised.
1. On the statistical point of
view, the logic and improvement of the way followed for the change-point search has
ensured a complete account of the information involved in the rank analysis of the series
and has avoided biases resulting from test misuses.
In particular, for testing
randomness against serial correlation, the circular and non circular rank serial
correlation test statistic involving both exactly the same information, they have thus the
same power efficiency. However, with its symmetric distribution close to the normal one,
the first test statistic avoids the complexity of distribution due to lack of symmetry of
the non circular one (Bartels 1982).
On the other hand, an efficient
answer has been given to the problem raised by Easterling and Peterson
(1995), concerning
the power limits of the methodology encountered when neglecting small discontinuities in
the series. In addition, when randomness is ascertained, testing the fit of any particular
distribution function, the normal one included, allows, with an increased power, further
investigations on the properties of the series.
For instance, though this was not
the case here, the detection of outliers in random groups with testing normality, makes
possible the determination of consecutive change-points separated by only one element.
2. As discontinuities may
eventually have a non climatic origin, verification of an eventual connection in the
acquisition data process has always to be made, though this operation has preferably to be
made before a climate instability investigation. The two phenomena being independent, the
effect may be either an amplifying or a damping one depending of their relative importance
(Sneyers, 1997, 1998, Sneyers et al., 1998).
3. The existence of a chaotic
evolution of the climate at all scales may theoretically be explained by the existence of
situations of indetermination in the day to day evolution of the General atmospheric
circulation (Sneyers 1997). It follows, that eventual man made effects on this evolution
may only introduce modifications in this evolution. As this effect may, here also, have
either an amplifying or a damping down effect, the problem of an eventual man made origin
(observational inhomogeneities included) reduces in each case to solving this question
with appropriate means.
