IMPROVED EXTREME WIND PREDICTION FOR THE UNITED-STATES

An investigation of sampling error due to short data records was investigated for the fastest mile of wind records used to form design wind speeds for the U.S. The data for 29 stations in the midwest were checked for statistical independence and for trend over the geographical area. A superstation was formed from the stations containing 924 station-years of record, and 10,000 years of record with the same Type I distribution were generated by random function generator. The simulated data was shown to have the same distribution of predicted 50-year speeds, if divided into 25-year long records, as occurred for the 29 real stations. The analysis showed that the major variation in predicted 50-year wind speed from station to station is the sampling error. It also showed a method for effectively removing most of the sampling error.