An improved method of extreme value analysis

Abstract

An improved method to estimate the probability of extreme events from independent observations is presented. The method, called VWLS, is based on minimizing the variance of order-ranked observations plotted according to their true probability and applying the least squares fitting. We show by Monte-Carlo simulations that this method provides estimates for the extremes that are considerably better than obtained by presently available EVA methods, particularly for small data sets. An additional benefit of VWLS is that its application requires no subjective methodological decisions by the user. Keywords: Extremes, Extreme value analysis, Statistical inference, Flood frequency analysis