Estimation of Extreme Wind Speeds with Very Long Return Periods

Abstract

Long series of hourly mean wind speeds and the maximum hourly 3-s gust are simulated for four sites in the British Isles in order to investigate methods for the determination of extreme wind speed events. The simulation is performed using a one-step Markov chain model. First, the observations are used to generate series of the same length as the real time series. It is found that the synthetic series reproduce well the means, standard deviations, and maximum and minimum wind speeds of the observed scales. As expected, they are less successful at reproducing the observation autocorrelation properties with the lagged correlation coefficients decaying two rapidly in the simulated series. The method is then used to generate synthetic series of 10, 50, 100, 1000, and 10 000 years in length. In a modified form of Monte Carlo analysis, each run of the Markov model is repeated 1000 times. The maximum wind speed from each run is extracted, and the mean of the 1000 values is taken to be the extreme wind speed for a return period equal to the length of the simulation. The results are compared with extreme wind speed calculated from conventional extreme event analysis. It is hypothesized that over the shorter return period (10 and 50 years) the comparison amounts to a validation since the conventional analysis may be expected to produce a reliable estimate of the true extreme. The authors find relatively close agreement between the extremes predicted by the two techniques at these shorter return periods. At the longer return periods (1000 and 10 000 years in length), the predictions from the synthetic time series are generally lower than those obtained from conventional analysis techniques. It is suggested that the results from the Markov model may be more realistic since one would expect them to be some theoretical maximum to wind speeds at any location imposed by atmospheric circulation characteristics.