The application of multivariate permutation methods based on distance functions in the earth sciences

Abstract

A variety of potential applications involving two classes of multivariate permutation methods based on distance functions are presented. These methods avoid the impossible distributional assumptions (knowing the analytic description of real data) associated with classical multivariate parametric methods. The relaxed distributional assumptions of permutation methods yield new statistical analyses which would be exceedingly difficult to formulate in terms of parametric methods. Applications of the first class include detection methods for various types of group difference, single and multiple clumping of rectilinear and cyclic phenomena, regular patterns, and first-order autoregressive occurrences (e.g., greenhouse effects). Applications of the second class include methods for evaluating numerical model output agreement with observed data and multivariate distribution-free linear-model analysis procedures.