2009 Volume 13 Issue 2 Pages 161-168
Any analysis of extreme values of wind climates with the aim of estimating design wind speeds suffers from statistical uncertainties induced by short observation periods. As can be shown by simulations, the identification of the ‘true’ parameters of a wind climate is virtually impossible. The situation is even worse if trends have to be evaluated in regard to their statistical significance. These trends may result from a global climate change which can also affect the wind climate. Particularly, trends in the number and intensity of storms seem to be of basic interest. The corresponding probability distributions are the Poisson distribution with describing parameter λ (average number of storms per year) and the Generalized Pareto distribution with describing parameters s and k (scale parameter and shape factor). The new approach is based on the idea that the observed parameters λobs, sobs, kobs and the corresponding linear trends aλ, obs and as, obs may be obtained from a large range of combinations of λ, s, k, aλ and as. In a first step, the probabilities are calculated that the observed parameters will be obtained for each combination λ, s, k, aλ and as. Then, for each combination, the design wind speed is calculated (for different target probabilities and different design working lives). Sorting the obtained design wind speeds and integrating their corresponding probabilities gives the non-exceedance probability curve of the probable design wind speed. The ‘best estimate’ is obtained by applying a target confidence interval of 75% which has previously been introduced as a basis for estimating design values for the resistance side when dealing with statistical uncertainties from confined ensembles. Based on the findings for the wind climate at Düsseldorf, it is strongly recommended to consider non-stationary features when specifying design wind speeds. The assumption of a stationary wind climate may lead to considerable underestimations of design wind speeds.
