Tail Length for Estimating Traffic Load Effects

Block maximum extreme value theory is commonly used for estimating lifetime maximum road traffic load effects. These effects originate from various loading situations, implying that block maxima may be drawn from different populations with distinct distributions. This poses a challenge when fitting extreme value distributions, as the events should be independent and identically distributed. To address this issue and isolate the critical load effect component, it has become standard practice to employ a censored fit to the upper tail of observed block maxima. The length of this tail significantly influences extrapolated load effects. However, determining the appropriate tail length remains uncertain. In this study, we investigate fitting to different tail lengths by simulating mixtures of extreme value distributions. Our findings show that an excessively short tail increases the statistical uncertainty of the fit, while an overly long tail risks introducing bias by incorporating events from non-critical load effect components. Consequently, the optimal tail length depends on both the sample size and the relative distributions of the different load effect components. This serves to discourage the use of certain accepted tail lengths without an assessment of the specific situation.