Aggregation of data from multiple recording stations for extreme wind analysis and prediction

Abstract

Accurate, reliable and robust techniques for the probabilistic estimation of extreme wind speeds are essential for the design of structures for wind loading. Aggregating gust wind data from various stations with similar, homogeneous wind climates into a ‘superstation’ for hazard analysis has been employed since the 1980′s to reduce the effects of sampling errors. A concern that has been raised recently is that prediction biases may arise from such aggregation, when the data exhibit non-homogeneity due to inevitable short data lengths or imperfect homogeneity of the wind climates. By Monte Carlo simulation, we show that superstation aggregation is an unbiased technique for high recurrence level estimations when an appropriate fitting method is used, and the apparent biases are dependent on the method used for fitting the hazard model. To ensure homogeneity, we introduce a de–trending technique for minimizing any biases in the aggregated wind data. Four model-fitting methods for superstation analysis are compared, and shown that the introduced de-trending method is effective for eliminating the biases due to sampling errors and non-homogeneity.