Consolidation of analysis methods for sub-annual extreme wind speeds

Abstract

This paper consolidates recent advances in methodologies in extreme-value analysis of wind speeds by using sub-annual maxima in conjunction with exact and penultimate extreme-value models. By avoiding asymptotic models and the associated issues of asymptotic convergence, the consolidated methodology is able to extend analysis further into the lower tail, greatly increasing the statistical confidence. The standard error in design predictions of dynamic pressure is reduced to less than a third of the corresponding standard error from annual maxima. The methodology is demonstrated by re-analysing wind speed data from previously published studies at sites in simple and in various mixed-mechanism climates.

Evaluating wind extremes in CMIP5 climate models

“…Statistical modeling of wind extremes remains an active research area (Harris 2005(Harris , 2009Makkonen 2008;Cook 2012Cook , 2014Makkonen et al 2013;Torrielli et al 2013). However, a number of analyses have shown that the annual maxima of wind speeds fit reasonably well to Gumbel distribution especially when wind speeds conform to the three-parameter Weibull distribution (Van den Brink and Können 2008; Pryor et al 2012a).…”

“…Estimates of extreme wind speeds at different return periods are usually computed using the Extreme Value Theory (Coles 2001). Several approaches have been suggested to compute quantiles at different return periods such as fitting annual maxima to Generalized Extreme Value (GEV) (Anastasiades and McSharry 2013) or Gumbel distribution, fitting the data over a threshold to Generalized Pareto Distribution (POT/GPD) (Holmes and Moriarty 1999), fitting r-largest values to GEV (Coles 2001), and extracting data using the Method of Independent Storm (IMIS or XIMIS) (Cook 1982;Harris 1999Harris , 2009) and building a Poisson process model (Cook 2014).…”

Impact of ASOS Real-Time Quality Control on Convective Gust Extremes in the USA

“…The XIMIS method [11], fitted by weighted least-mean-squares (wLMS) and displayed on Gumbel axes, was preferred [12] over other extreme-value analysis (EVA) methods for the following reasons:…”

“…The accuracy of ( 5) and ( 6) is assessed in [12]. The distributions of plotting position around the mean, y m , are also the distributions of the m-th highest values for repeated samples of the observation period, R. In practice, as there is only one observation period, the distributions of the m-th highest values are used to assess the confidence that can be placed in the individual values of that one sample.…”

“…All the observations contribute to the fitted model, inversely weighted by their variance, so the model confidence limits are much tighter. Model confidence can be evaluated numerically by the "jackknife" or by the bootstrap [16], but this is outside the scope of the present study and is addressed elsewhere [12].…”