Figure 5

Data subsets with homogeneous pressure approximate Poissonian statistics. High-wind velocity persistence statistics, \(v\ge 12\,{\rm{m}}/{\rm{s}}\), is analyzed at Alpha Ventus, based on the downscaled ERA-Interim data from 1980–2010⁴⁶. Three different data sets are compared: First, the original data, consisting of 31 years of measurements is split into 31 equally sized data sets, based on the year it was recorded (Data: Year). Alternatively, the data is split based on approximately homogeneous f-parameter (Data: f-parameter). Finally, this is compared to an artificial Poissonian process with return times as estimated from the exponential distribution, generating an equal number of data points (Poisson). The q-values of the full sets are indicated by colored lines at the sides, both for the real data as well as the Poissonian process. The full data q-value is larger than the q-values of most subsets. Furthermore, splitting the data arbitrarily according to calendar years leads to more values at large q than if the data is conditioned on the f-parameter. Conditioning on the f-parameter approximates the Poisson distribution much better than yearly conditioning, when computing the Wasserstein distance⁶⁴ of the distributions. The box plot gives the median as a black line, the 25% to 75% quartile as a yellow box and minimum and maximum value as the whiskers.