Global decadal-to-multidecadal climate variability and its evolution mechanisms: a review and outlook

Decadal-to-multidecadal climate variability (DMCV), considered here over periods of approximately 8-80 years, is a major source of regional climate predictability on societally relevant timescales. This review synthesizes the spatial structures, temporal characteristics, underlying dynamics, and climatic impacts of the principal low-frequency modes in the North Atlantic, North Pacific, and Southern Hemisphere. Internal variability provides the organizing framework, while the roles of anthropogenic aerosols, greenhouse gases, volcanic forcing, and other external drivers are explicitly considered. Particular emphasis is placed on how atmospheric bridges, including stationary Rossby-wave trains and adjustments of the Walker circulation, interact with oceanic pathways involving the Atlantic Meridional Overturning Circulation, the Indonesian Throughflow, and the global overturning circulation to transmit and transform signals across ocean basins. We further assess the extent to which observed spectral peaks represent dynamically constrained, quasi-oscillatory behavior rather than stochastic atmospheric forcing integrated by ocean memory. Evidence from observations, paleoclimate reconstructions, CMIP5/6 simulations, large ensembles, and pacemaker experiments is evaluated alongside diagnostic approaches such as empirical orthogonal function analysis, multichannel singular spectrum analysis, and dynamic mode decomposition. Current evidence indicates that low-frequency modes modulate the background probability of regional hydroclimatic anomalies and climate extremes, but contemporaneous correlations and phase composites alone do not establish causality. Major uncertainties arise from the limited length and inhomogeneity of instrumental records, sparse direct observations of subsurface circulation, model-dependent amplitudes and persistence, and incomplete separation of internally generated variability from forced responses. Priority directions include coordinated single-forcing and sensitivity ensembles, sustained ocean observations, millennial-scale multiproxy reconstructions, eddy-resolving simulations, physically constrained causal inference and machine learning, and out-of-sample evaluation of decadal predictions across regions, variables, and lead times.