Mixture distributions in climatological research

Description/Abstract

Mixture distributions have had a long history in statistics where the central problem has been the decomposition of multiple distributions into their component parts and the estimation of the parameters of these sub-distributions. Here, in the climatological context however, the nature of these complex distributions is examined, without any attempt to decompose them, examining not only data distributions for a range of climatological variables, but their anomaly distributions as well. Histograms are easily generated, providing a rich visual display for intercomparison purposes. They serve to readily flag suspect or unusual data and can be suggestive of underlying mechanism. They can provide information which can be of considerable value for extreme value studies. Finally, an enormously rich statistical literature exists on the nature of distributions, physical processes leading to their development, and techniques for parameter estimation. Of particular relevance to model/data intercomparisons, there exist a broad range of tools for addressing the fundamental question: ``are two distributions statistically the same?``.