Page 15 - Lloreta_alii_2001

DISCUSSION

Long-term climate forcing

Whereas NAO displayed short- to long-term variability (being different among

the 3 indices), spectra of LOD, temperature and BFT time series were similar and

largely dominated by low-frequency signals. Analyses in time domain did not further

display any clear relationships between BFT long-term fluctuations and NAO. The

results of the regression and correlation between BFT time series and LOD index were

inconsistent, but they were clear and consistent between BFT time series and

temperature time series: BFT long-term fluctuations appeared closely and negatively

related to those in temperature.

We stressed that combining many variables may affect the level of significance of

the statistical tests and, thus, the confidence in the above results. However, the Holm’s

procedure, which was applied to correct for multiple testing, confirmed the above

findings, so that the conclusions may be considered as robust. Another difficulty relates

to serial correlation in time series, which decreases the effective degrees of freedom

(Bartlett, 1946). Analyses showed that regressions generally vanished when

autocorrelation was taken into account, indicating that significance was due to the

autocorrelated signal, i.e. the long-term fluctuations. Therefore, we may conclude that

significant relationships between BFT time series and temperature time series only

come from opposite long-term fluctuations, but standard statistical analyses do not

allow assessment of the real level of significance of such relationships. A last

methodological point relates to the fact that the analyses have been performed on

different time periods (simply because the BFT time series occur over different periods

and have various lengths). These differences should not bias the comparisons, especially

if we assume stationarity (which is the case for most of the series). Any statistical

16