Locally Stationary Analysis of Hurricane Windfields

Locally stationary analysis considers signal models with time-varying mean and spectrum dependent on an unknown time-dependent parameter. The appropriateness of this model for wind analysis is determined by the relation between decorrelation length and length of (approximate) stationarity. Some nonstationary wind fields appear to evolve slowly enough to easily meet this requirement. Hurricane passage often provides an example. Spectral estimates for boundary layer wind fields have, in the past, used reduced frequency, in which frequency is normalized by the mean wind speed. This and Taylor's hypothesis in essence change the signal representation from time-domain to space-domain. Higher mean wind speed simply moves turbulence of the same spatial scale past the anemometer more rapidly. This model and locally-stationary estimates for normalized spectrum, based on locally transforming the time-scale, provide an analysis benchmark for establishing confidence intervals and data requirements for detecting changing spectral characteristics in extreme windfields. The techniques are demonstrated using wind time-series collected by Texas Tech’s Wind Engineering Mobile Instrumented Tower Experiment (WEMITE) during Hurricane Bonnie.