Page 12 - Lloreta_alii_2001

indicated that long-term fluctuations in BFT time series were not statistically related to

the NAO.

BFT and LOD

The decomposition in frequency domain of the LOD time series showed a red-

-1
shifted spectrum (frequencies < 0.1 yr ), similar to those of the BFT time series

(Fig. 4). Five of the eight regressions between trap catches and LOD were significant

(Table 3), but the boxplot of the p-values exhibited a rather large distribution of the

probabilities (Fig. 5b). Further, the slopes were low (either positive or negative) and

both positive and negative relationships appeared significant (3 and 2 respectively,

Table 3). More careful inspection revealed strong structure in the residuals and periods

of positive relationships (1750-1810, 1920-1960), alternating with periods of low or

negative relationships (1850-1890). Fits of the GLS models all led to positive and non-

significant regressions, indicating that long-term trends were the most common feature

between trap catches and LOD. 75% of the correlation coefficients were significant and

about 60% when correction for multiple testing was applied. Thus, BFT time series and

LOD showed fairly similar patterns in frequency domain, but regressions were not

consistent among all the analyses and did not indicate any clear relationship.

BFT time series and temperature

Spectra of the two proxies of NH temperature were dominated by low frequencies

and presented red-shifted spectra, similar to those of the BFT time series (Fig. 4). All

the regressions between trap catches and NH temperature displayed strong negative

slopes (from -2.90 to -0.17, Table 3). All those with the Jones’ proxy were significant at

the 1% level (Table 3, Fig. 5c) and exhibited normally distributed residuals. Concerning

D’Arrigo’s proxy, two BFT time series (i.e. Bonagia and Favignana) were not

13