A Description of the Madden–Julian Oscillation Based on a Self-Organizing Map

Abstract

Abstract This study introduces a nonlinear clustering technique based on a self-organizing map (SOM) algorithm to identify horizontal and vertical structures of the Madden–Julian oscillation (MJO) through its life cycle. The SOM description of the MJO does not need intraseasonal bandpass filtering or selection of leading modes. MJO phases are defined by SOM based on state similarities in chosen variables. Spatial patterns of rainfall-related variables in a given MJO phase defined by SOM are distinct from those in other phases. The structural evolution of the MJO derived from SOM agrees with those from other methods in certain aspects and differs in others. SOM reveals that the dominant longitudinal structure in the diabatic heating and related fields of the MJO is a dipole or tripole pattern with a zonal scale close to that of zonal wavenumber 2, as opposed to zonal wavenumber 1 suggested by other methods. Results from SOM suggest that the MJO life cycle may be composed of quasi-stationary stages of strong, coherent spatial patterns with relatively fast transition in between that is less coherent and weaker. The utility of SOM to isolate signals of an individual MJO event in a case study is illustrated. The results from this study show that some known gross features of the MJO are independent of diagnostic methods, but other properties of the MJO may be sensitive to the choice of diagnosis method.