|
Authors: Jonathan F. Donges, Norbert Marwan, Yong Zou and Jürgen Kurths
Climate networks present a promising novel tool for the statistical analysis of multivariate climatological data sets. Each spatial grid point in the climate data set is identified with a vertex of the corresponding climate network, and edges are introduced between pairs of grid points with statistically interrelated dynamics as measured by, e.g., Pearson correlation or mutual information. Due to the effect of noise, the transitivity effect and other issues, this construction method contains an intrinsic uncertainty on the network structure. Given this uncertainty, it is very important to evaluate the statistical significance of measured network properties such as clustering coefficient, average path length, degree distribution or various vertex centrality fields with respect to a given null hypothesis. Here we present significance tests based (i) on network surrogates (Erdös-Rényi and configuration model) and (ii) on networks constructed from surrogate data sets using time series surrogates (shuffled, Fourier and twin surrogates). We demonstrate the proposed significance tests for a global climate network constructed from coupled model surface air temperature data. |
|