Design Wind Speed Prediction

Abstract

Structures are designed with the intention of safely withstanding ordinary and extreme loads over the entire intended economic lifetime. Designation of appropriate design levels is especially difficult for natural processes, for which many decades of reliable data are desirable, but is often basically unattainable. One such area of interest is the prediction of extreme wind speeds. Recently, multiyear, mechanically recorded data from five sites in the Pacific Northwest were obtained. The data consist of hourly 1-s maximum gust speeds, wind direction, temperature, and barometric pressure. Using these variables, the data are examined by season to determine statistical characteristics, including their dependencies. In particular, histograms of the wind gusts are examined in detail at each location to determine whether they can be modeled by a single probability distribution. Statistical time series analysis is traditionally dependent on the assumption that all observations arise from a single distribution, but recent theoretical advances have been made with mixed distributions in extreme value theory. The suitability of the observed wind data for analysis using mixed distributions is examined, and the peaks-over-threshold technique is applied. The results are compared to those obtained by assuming the data arise from a single distribution. Rather striking results indicate the importance of the mixed distribution approach, including seasonal and direction variations. We also note that the theory of mixed distributions in extreme value theory can be applied to any data for which extremal computations are desired.