The relative merits of three contrasting empirical orthogonal decomposition methods in common use (namely, Proper Orthogonal Decomposition, Biorthogonal Decomposition and Multivariate Singular Systems Analysis) are considered as applied to baroclinic flow data. The regimes analysed are a steady, drifting wave, a modulated amplitude vacillating wave flow and a neighbouring multi-mode state which exhibits intermittency. The results are used to make a qualitative comparison of the methods in terms of convergence properties, variance capture and eigenfunction structure. The feasibility of using the resulting empirical orthogonal functions to transform partial differential equations to ordinary differential equations by Galerkin projection is mentioned. © 1997 Elsevier Science B.V.
