The accuracy of design values predicted from extreme value analysis

This paper reports a study of predicted values of the 1 : 50 yr windspeed produced by four different methods of extreme value analysis. The methods used were the obsolete Classical Gumbel method; Gumbel–Lieblein BLUE; the Modified Gumbel method and the Method of Independent Storms. For each method, synthetic wind records were generated consisting of 100 storm maxima for each year and for a duration of 10(10)50 years. The parent distribution was chosen so that q=V2 had an exponential probability distribution. For each method tested, 10 000 data sets were analysed and the slope and intercept parameters were used to predict a value of q50 for each set. The 10 000 realisations of q50 were then examined for bias and statistical variation about the (known) theoretical value. The tests indicate that where sufficiently continuous data is available, the Method of Independent Storms should be used. Otherwise if only annual maxima are available, Gumbel–Lieblein BLUE gives the best results for small data sets, but Modified Gumbel is an acceptable alternative, particularly for samples N>25.