Non-stationary frequency analysis of typhoon-induced wind and wave extremes for China’s offshore regions

Many environmental variables exhibit non-stationary characteristics under global warming, yet such assessments for typhoon-induced multi-hazards remain limited. This study develops an integrated non-stationary joint probability modeling framework that incorporates trend detection, cyclicity analysis, non-stationary marginal extreme value modeling, and multivariate dynamic dependence evaluation. A novel implementation of the Hilbert-Huang transform is proposed to capture the cyclic behavior of constituent variables, potentially caused by various atmospheric circulations, for improved physical realism. Based on the use of ERA5 reanalysis data for China’s offshore regions, the results show that over 40% of sites exhibit increasing trends, and approximately 25% show declining variability and weakening wind-wave dependence. The cyclic effects associated with the El Niño-Southern Oscillation are identified, with a dominant period of about 3.2 years. The generalized additive model-based non-stationary models can better capture nonlinear features compared to linear and stationary models. The variability of time-dependent design loads is mainly governed by non-stationary marginal models rather than by copula-based dependence models. Notably, lifetime exceedance probabilities may be higher under non-stationary conditions compared to those obtained under stationary assumptions. These findings emphasize the importance of non-stationary analysis and modeling of typhoon-induced wind and wave extremes for better-informed offshore structural design and risk adaptation in a changing climate.