COMMENTS ON - THE EFFECT OF DOMAIN SHAPE ON PRINCIPAL COMPONENTS ANALYSES

In a recent article claiming to deal with the effect of domain shape on principal component analysis, Legates attempts to use previously published research and analyses of two exploratory data sets to try to dismiss Buell's work on the topography of empirical orthogonal functions. Legates' logic and data sets suffer from a number of flaws, including omission of key results from Buell, neglect of dozens of published papers discussing domain dependence, misinterpreting and misrepresenting numerous references to Richman, and using data sets with inhomogeneities between land and water as well as very low temporal degrees of freedom. This Comment addresses some of these shortcomings by exploring what other researchers have concluded about Buell patterns and how they have minimized the problem. It also serves to provide a proper interpretation of PC loading fields consistent with Richman, and to point out several potential problems in the two data sets used by Legates, which might lead to inferior results.