Statistical evaluation of ECMWF extreme wind forecasts
Haiden, Thomas ; Magnusson, Linus ; Richardson, David
Here we evaluate the skill of the ECMWF forecasting system in predicting high wind events over a large sample. Events can be defined based on absolute thresholds (e.g. gale-force winds) or the degree of severity compared to climatology (e.g. wind speeds above the 99th percentile). While the absolute value may be more relevant with respect to damage, the percentile-based definition is useful for producing spatially or seasonally aggregated scores, since by definition the number of events becomes comparable between different regions and seasons. An additional reason for choosing a percentile threshold is that the actual impact of an event of given absolute intensity in a certain region will depend on how often it occurs in that location, as this will influence the degree to which the natural environment, buildings and infrastructure are adapted to it. In any case, the choice of specific thresholds involves a compromise. A high threshold is more targeted to rare events but at the cost of a small sample, while a low threshold may provide more reliable statistics but fails to distinguish the skill in forecasting extreme weather from the more general skill of the forecast. By verifying wind speed forecasts against SYNOP observations, we will show that predictions of severe wind events have benefited from improvements in the forecasting system as much as more ‘normal’ weather as suggested by improvements in standard skill scores.
Accès à la notice sur le site du portail documentaire de Météo-France