The Proper Orthogonal Decomposition (POD) is a procedure to compute an orth
ogonal basis from a time series of spatial fields. This basis is optimal am
ong all linear decompositions, in the sense that for a given number of mode
s, the projection of the original signal onto the subspace will contain the
most variance on average. This algorithm is applied to streamfunction fiel
ds derived from measurements of the flow in the thermally forced rotating a
nnulus experiment. Results of this analysis are presented, and a method to
derive low-dimensional models of the flow by projecting the equations of mo
tion onto these empirical eigenfunctions is discussed. (C) 1999 Elsevier Sc
ience Ltd. All rights reserved.