It should be here recognized that the confidence intervals shown in Figure 2 are most likely too narrow since they assume normally distributed data and independent years.
It should be also emphasized that the presence of interdecadal modulation in actual or hypothetical teleconnections is ubiquitous and can arise from stochastic noise.
Following the Monte Carlo technique proposed by Gershunov et al. [2001], the standard deviation of the running correlations shown in Figure 2b can be compared to the range of standard deviations which is obtained by simulating N = 5000 sequences of running correlations between pairs of correlated white noise time series.
Using the overall correlation coefficient between the SCI anomalies and each of the selected dynamical index estimated from the full record (111 years), we find that the 95th percentile ofthe standard deviation is comprised between 0,20 and 0,23 depending on the selected dynamical index. The actual standard deviation of the time series shown in Figure 2b is comprised between 0,15 (for the
AO index)and 0,22 (for the EOF-derived
PNA index) and is therefore weaker than might be expected by chance.
This result means that the recent strenghtening of the snow-AO anticorrelation might be a purely stochastic variation.In other words, any empirical prediction of the winter (N)AO based on the recent anticorrelation withthe Siberian snow cover in fall will be obviously overconfident
Segnalibri