Tail-dependence-based up-crossing analysis for extreme value prediction

Extreme value prediction based on up-crossing analysis is widely used for reliability assessment of engineering structures. However, existing approaches for non-Gaussian processes often rely on Gaussian dependence assumptions through memoryless translation models, which may be inadequate when the upper-tail dependence between consecutive observations differs from that of a Gaussian copula. This study develops a tail-dependence-based up-crossing framework for extreme value prediction of non-Gaussian dependent random sequences. The Ledford–Tawn representation is incorporated to characterize the joint upper-tail behavior of consecutive observations, leading to an explicit formulation of the up-crossing rate in terms of the marginal distribution and tail dependence parameters. A sensitivity analysis is further conducted to quantify how variations in tail dependence parameters affect return-level estimates. Numerical examples and measured wind speed records demonstrate that the proposed method accurately reproduces up-crossing rates and extreme-value exceedance probabilities, while revealing the potential bias caused by inappropriate Gaussian dependence assumptions.