Eurocode standard recommends using fundamental basic wind velocity (characteristic velocity) as the design value in civil engineering. There are different approaches to estimate this value depending on the climate features of the given area and the quality of environmental data. The estimation of the characteristic value requires statistical analysis of historical data regarding wind velocities measured throughout the country at meteorological stations. The results of the analysis are probability density distributions of this random variable for each meteorological station. On this basis, values of characteristic wind velocity with a mean return period of 50 years are determined. The zones with uniform velocities are delineated on the map of the country. In the case of Poland the last evaluation of wind zones took place over 15 years ago. Higher quality of measurement data on the one hand, and the introduction of the second generation of Eurocode standards on the other hand, create a need to check and update these zones. This work presents theoretical basis for the estimation of characteristic values of random variables in the context of wind velocity, comprehensively reviews practical methods used for this purpose and summarizes current situation in Poland, finally discusses the issues related to the heterogeneity of wind data, illustrating them with an example.

The paper deals with application of the Gumbel model to evaluation of the environmental loads. According to recommendations of Eurocodes, the conventional method of determining return period and characteristic values of loads utilizes the theory of extremes and implicitly assumes that the cumulative distribution function of the annual or other basic period extremes is the Gumbel distribution. However, the extreme value theory shows that the distribution of extremes asymptotically approaches the Gumbel distribution when the number of independent observations in each observation period from which the maximum is abstracted increases to infinity. Results of calculations based on simulation show that in practice the rate of convergence is very slow and significantly depends on the type of parent results distribution, values of coefficient of variation, and number of observation periods. In this connection, a straightforward purely empirical method based on fitting a curve to the observed extremes is suggested.