Authors: H Van de Vyver A W Delcloo
Publish Date: 2011/01/27
Volume: 105, Issue: 3-4, Pages: 417-429
Abstract
Generalized Pareto distributions GPD are frequently applied for the statistical analysis of extreme wind speeds A central topic in extremevalue theory is the adaptive estimation of the extremevalue index γ Several authors have demonstrated a high sensitivity of γ against the threshold when analyzing extreme wind speeds This undesirable effect introduces the difficulty to provide reliable quantile estimates This paper aims to bring this problem to meteorologists and proposes a stable estimator the Zipf estimator for γ This could allow a more objective prior identification of the sign and range of γ The method is based on regression in the socalled generalized quantile plots A comparative study with a classical estimator the probabilityweighted method is made and it is shown that the Zipf estimator significantly decreases the variance in the calibration of the GPD to extreme wind gusts Finally the new methodology is applied to get improved prediction of extreme wind gusts in Belgium
Keywords: