Estimating Uncertainty in the Extreme Value Analysis of Data Generated by a Hurricane Simulation Model
S Coles, Emil Simiu
Extreme value analyses of any environmental phenomenon are fraught with difficulties, but the additional difficulty of collecting reliable data during hurricane events makes their analysis even more complicated. A widely accepted procedure is to use calibrated hurricane models for the simulation of realistic hurricane events, whose data can be subjected to standard extreme value procedures. The estimation uncertainties which arise from such analyses depend upon (1) the extent to which the hurricane models are physically realistic, (2) the length of the simulated series, which consists of shout 1,000 or even 10,000 simulated events, and therefore introduces negligible errors, and (3) the length of historical record on which the siulated records are based, which usually consists of about 50 events. In this note we propose the use of resampling schemes as an attempting to obtain some reasonable measure of uncertainties due to the relatively short length of the historical record. An intuitive, na ve procedure is first described, which leads to an alternative approach that has connections with the statistical procedure of bootstrapping. Standard application of these procedures for extermes induces bias, and we propose a simple, though non-standrd method for reducing this effect. The results are illustrated on a dataset of simulated hurricane wind speeds for an eastern US coastline location.
Journal of Engineering Mechanics-Asce
bootstrap, Building Technology, extremes, hurricane wind speeds, simulation, wind engineering