Page 13 - Lloreta_alii_2001
P. 13

significantly related to it and further displayed time-dependent structure in the residuals.

                  However, the 7 other regressions were significant at the 5% level (Table 3) and


                  displayed normally distributed residuals. The boxplot of the probabilities tended to

                  confirm that BFT time series were, in general, significantly related to D’Arrigo’s time


                  series, but this was less obvious than with Jones’ time series (Fig. 5c). Most of the fits

                  of the GLS model led to lower slope coefficients (from -2.76 to 0.86, Table 3) and to a


                  significant drop in the strength of the relationship (only 2 of the 14 regressions

                  remained significant, Table 3). This again shows that long-term trends were the most

                  common feature to BFT, Jones’ and D’Arrigo’s time series. These findings were


                  confirmed by the highly negative correlation coefficients, among which 100% (for

                  Jones’ proxy) and about 80% (for D’Arrigo’s proxy) were significant at the 5% level for


                  both corrected and non-corrected correlation analyses (Table 3).

                        The spectra of the four local temperature time series were also dominated by long-


                  term variability, being highly similar to those of the BFT time series (Fig. 4). 23

                  regressions between BFT and local temperature were significant at the 1% level and the

                    th
                  24  one at the 5% level (Table 3). The boxplots of the probabilities displayed

                  unambiguous results for all the four temperature time series (Fig. 5d). All regressions


                  exhibited downward slopes (among which 12 were particularly strong, with slope < -

                  0.5, Table 3). The GLS regressions led to  lower and non-significant relationships in

                  86% of the cases and weak slopes coefficients (from -0.30 to 0.26, Table 3). This again


                  indicated that long-term signals were common to BFT and local temperature time series.

                  Finally, correlation analyses confirmed the strong and negative relationships between


                  BFT and local temperature. All the correlation coefficients were significant at 5% (and

                  for the most at 1%), even when correction for multiple testing was applied.









                                                            14
   8   9   10   11   12   13   14   15   16   17   18