Fig. 3
From: Global-scale multidecadal variability missing in state-of-the-art climate models
M-SSA analysis of observed and model-simulated internal secular variability. a M-SSA spectra showing the levels of variability associated with dominant space–time M-SSA patterns underlying the data. Results for differences between the observed and CMIP5-simulated secular signals (observed internal variability): ensemble-mean spectrum and standard uncertainty (over the 111 estimates), black curves and error bars, respectively; analogous results for deviations of individual-model secular signals from each model’s ensemble mean (simulated internal variability), blue curves and error bars; 99th percentile of simulated spectra, blue dashed curve; the 99.99th percentile of variances obtained by projecting the simulated internal variability onto the M-SSA patterns of observed internal variability, red curve. b Locations of regional SAT indices. c Reconstructed time series associated with the leading M-SSA pair of observed–model-simulated data differences (observed internal variability) in select regional indices: NA North Atlantic, NP North Pacific, SWP Southwest Pacific, AA Antarctica, SA South Atlantic, A Arctic; Global Multidecadal Oscillation (GMO) time series represents the reconstruction of the global-mean temperature. All of the time series are normalised to unit standard deviation; the actual standard deviations of A and AA indices are around 0.6 °C; that of all others—0.1 °C
