Fig. 5

Measures of similarity between various characteristics of the observed and CMIP5 simulated NINO3.4 time series. Shown are ensemble averages of these characteristics over multiple runs of individual models (see the model acronyms on the left). The first two columns compare the observed and simulated wavelet spectra, using the root-mean-square (rms) distance and Pearson correlation, both computed in the frequency space. Next three pairs of columns display measures of similarity between the observed and simulated interaction maps analogous to the observed maps of Fig. 1, namely the phase synchronization, phase–phase causality and phase–amplitude causality maps. Each pair presents two distinct similarity measures: the Pearson correlation (corr) in the phase space of the corresponding map, as well as the Adjusted Rand Index (ARI; see text)