Changes in extreme wind speeds in NW Europe simulated by generalized linear models

We investigate the capability of generalized linear models (GLMs) to simulate sequences of daily maximum wind speed (DMWS), at a selection of locations in NW Europe. Models involving both the gamma and Weibull distributions have been fitted to the NCEP reanalysis data for the period 1958-1998. In simulations, these models successfully reproduce the observed increasing trends up to 0.3 m/s per decade in coastal or oceanic locations for the wintertime and the decreasing trends down to -0.2 m/s per decade in inland Europe for the summertime. Annually extreme winds exhibit an increasing tendency (with median estimates up to 0.6 m/s per decade) at the studied locations. The gamma model slightly overestimates the upper percentiles of the wind speed distribution, but reproduces trends better than the Weibull model. In both the NCEP data and GLM simulations, local extreme DMWS events (defined in terms of threshold exceedances) have increased dramatically in frequency during winter; decreasing trends are more common in summer. The NCEP data indicate similar trends in the frequencies of large-scale windy events (defined via simultaneous exceedances at 2 or more locations). Overall, these events have increased in number; at the scale of the North Sea basin, their number may have changed from 3-5 days per year during the earlier decades, to 5-7 days per year during later decades based on observational estimates. An increase in the frequency of large-scale extreme winter storms is implied. The GLMs underestimate these large-scale event frequencies, and provide imprecise estimates of the corresponding secular trends. These problems could be rectified by using a better representation of spatial dependence. The present results suggest that GLMs offer a useful tool to study local climate extremes in the context of changing climate distributions; they also provide some pointers towards improving the representation of extremes at a regional scale.