Getting the most from your extreme wind data: A step by step guide

Abstract

Models for extremes of environmental processes have been studied extensively in recent years. The particular problems arising when attempting to estimate return levels from sequences of measurements on the appropriate variables have been considered in some detail. In particular, the aspects of seasonal variation and short-range dependence have received a great deal of attention. In this paper we present a case study based on 10 years of hourly wind speed measurements collected at a U.K. site, elucidating the most successful procedure emerging from an extensive study of this data. The basic model (in which an extreme value distribution is fitted to cluster peak excesses over a high threshold) is standard. However the emphasis is on a number of practical problems which will arise when such models are fitted to wind speeds, but which have received little consideration. These include: model selection and assessment of model adequacy when the threshold, and some or all of the parameters, are allowed to vary seasonally; the choice of the best combination of threshold and cluster identification procedure; and the choice of a measure of precision for return level estimates. The aim is to suggest an algorithm which can be generally applied to the problem of gust return level estimation at individual sites.