A novel method for the detection of persistent and recurrent climatic regimes as almost invariants of the transfer operator

A novel method and its numerical implementation to find almost invariants of a dynamical system will be presented, with applications to the detection of persistent and recurrent climatic regimes, coherent structures in ocean flows and spatial patterns of climate variability. The method is based on an estimation of the transfer operator of the particular dynamical system. The detection of almost invariants is posed as a Markov reduction problem with a minimization of the relative entropy, here a measure of the distance between the fine-grained system and the reduced Markov chain. It is implemented using a fast-greedy algorithm from complex network theory.

Two applications in different domains of climate science are presented. In the first one, two persistent and recurrent atmospheric flow regimes are identified from a simulation of a barotropic model of the northern hemispheric atmosphere with realistic winter forcing. The regimes correspond to the well-known blocked and zonal circulation regimes of the northern hemisphere. Secondly, the algorithm is applied to a correlation network estimated from 140 years of sea surface temperature data to identify spatial patterns of variability. Dominant patterns on interannual to decadal time-scales are found in the tropical Pacific (El Niño-Southern Oscillation), the North Atlantic (the Atlantic Multidecadal Oscillation) and the Indian ocean and West Pacific.