Page 11 - Lloreta_alii_2001

RESULTS

BFT and NAO

As already depicted by Ravier and Fromentin (2001), spectra of BFT time series

-1
were all strongly dominated by low frequencies (< 0.1 yr , i.e. periods > 10 yrs); short-

to medium-term variations being weak or negligible (Fig. 4). In contrast, no common

patterns emerged from the 3 NAO spectra. The instrumental Hurrell’s index presented a

-1
broad band spectrum dominated by short-term variability (> 0.4 yr ), with two smaller

-1
peaks at medium and low frequencies (around 0.3 and 0.05 yr ). Long-term signal
emerged from the longest NAO time series. The Cook’s index also displayed a broad

-1
band spectrum with a peak at low frequencies (< 0.1 yr ) and secondarily at around
-1
-1
0.25 and 0.5 yr . Appenzeller’s index is dominated by low frequencies < 0.2 yr (i.e.
-1
periods > 5 yrs), with two peaks at around 0.02 and 0.15 yr . So, in contrast with BFT
spectra, medium- to short-term variability also appeared for all the three NAO indices.

Pairwise regression analyses between Hurrell’s index and BFT time series led to non-

significant relationships, as it is showed by the boxplot of the p-values of the six

regressions (Fig. 5a). The slopes were always weak (ranging from -0.01 to 0.08,

Table 3). It is worth noting that an increasing length of the series (i.e. analyses with the

2 NAO proxies) did not lead to stronger relationships (Fig. 5a): most of the regressions

also appeared non-significant and the slope coefficients were low (and either positive or

negative, i.e., from 0.32 to 0.07, Table 3). The results of the GLS models (that took

autocorrelation into account) were similar to the linear regressions and always non-

significant at 5% (Table 3). Correlation analysis confirmed the previous findings: 72%

of the correlation coefficients were non-significant; this proportion rose to 88% when

correction for multiple testing was applied. The results of all the above analyses

12