J3.8 FINDING INTERESTING CLIMATE PHENOMENA BY EXPLORATORY STATISTICAL TECHNIQUES

A. Ilin,Harri Valpola,E. Oja
Abstract:Finding a meaningful representation of the observed data is a central problem in many research fields. In many applications, collected data contain measurements of a set of variables xi obtained at different time instances t. These measurements are typically represented as a set of vectors x(t) = [ x1(t) . . . xi(t) . . . xN (t) ] T collected during the observation period t = 1, ..., T . In climatology, such datasets may contain observed weather conditions in certain spatial locations during a certain period of time. A typical task of statistical data analysis is to find important, interesting data characteristics. Often, these interesting properties are captured by a set of vectors s(t) = [ s1(t) . . . sj(t) . . . sM (t) ] T obtained by a linear transformation of the observed data:
What problem does this paper attempt to address?