Extreme wind estimates for the Eurocode: a study into the shape of the distribution of extreme wind speeds

The research contributes to reassessing the wind climatology underlying the Eurocode construction standard, which ensures building safety over long lifespans. The current Eurocode is based on only fifteen years of wind data fitted to a Gumbel distribution, leading to high uncertainty in estimates of extreme winds. More accurate estimates are crucial, as underestimation increases the chance of structural failure, while overestimation increases construction costs.
The project, a collaboration between KNMI, TNO, and the University of Southern Denmark, builds on the method of De Valk and Van den Brink (2025), which uses measurements and climate model data to extrapolate extremes via Generalised Extreme Value (GEV) distributions. These distributions are described by a location, scale, and shape parameter, the latter controlling the behaviour of the distribution’s tail and thus determining the likelihood of very extreme events.

This study investigates how significant the differences are in shape parameters for extreme wind speeds between regions (the Netherlands and Europe) and between different (climate) models (RACMO, SEAS5, and CORDEX), and whether data transformation can help reduce these differences.

Previous work by Tribut (2024) using the RACMO regional climate model showed substantial differences in shape parameters between two Dutch sites (Schiphol and Cabauw). This raised questions about spatial variations in the parameter and the possible influence of different climate models. In this study, three climate models—RACMO, CORDEX, and SEAS5—were analysed for Europe and the Netherlands, focusing on five Dutch sites (Cabauw, Schiphol, Eelde, Eindhoven, and K13) with the longest and most homogeneous measurements. The effects of data averaging (hourly versus ten-minute) and data transformation were also examined.

In this study the annual maximum wind speeds of the (climate) models are fitted to a GEV distribution instead of the method that more advanced method that Tribut (2024) suggested (fitting Peak of Threshold data to a Generalised Weibull GW distribution) as this method reduced the thousands of years of climate model data to a manageable size. To improve the convergence of parameter estimates, data were transformed by raising them to the power of 1.75, after which new GEV fits were computed. This transformation reduced differences in shape parameter between both regions and models, producing more uniform shape parameters and consistent 10,000-year wind speed estimates.

The findings show that differences between climate models are not statistically significant for the Netherlands or most of Western Europe. Hence, model choice does not strongly affect extreme wind estimation. However, data transformation is recocommended, as it reduces regional variability and improves parameter convergence.