Extreme events in the Mediterranean area: A mixed deterministic-statistical approach

Statistical inference suffers for severe limitations when applied to extreme meteo-climatic events. A fundamental theorem proposes a constructive theory for a “universal” distribution law (the Generalized Extreme Value distribution) of extremes. Use of this theorem and of its derivations is nowadays quite common. However, when applying it, the selected events should be real extremes. In practical applications a major source of errors is the fact that there is no strict criterion for selecting extremes and, in order to “fatten” the statistical sample very “mild” selection criteria are often used. The theorem in question applies to stationary processes. When a trend is introduced, inference becomes even more problematic. Experience shows that any available a priori knowledge concerning the system can play a fundamental role in the analysis, in particular if it lowers the dimensionality of the parameter space to be explored. The inference procedures serve, then, the purpose of testing the reliability of inductive hypothesis, rather than proving them. Within the above general context, analysis of the hypothesis that the frequency and/or intensity of extreme weather events in the Mediterranean area may be changing is proposed. The analysis is based on a combined deterministic-statistical approach: dynamical analysis of intense perturbations is combined with statistical techniques in order to try to formulate the problem in such a way that meaningful conclusion may be achieved