Uncertainty quantification in estimation of extreme environments

Abstract

We estimate uncertainties in ocean engineering design values due to imperfect knowledge of the ocean environment from physical models and observations, using Bayesian uncertainty analysis. Statistical emulators provide computationally efficient approximations to physical wind–wave environment (i.e. “hindcast”) simulators and characterise simulator uncertainty. Discrepancy models describe differences between hindcast simulator outputs and the true wave environment, where the only measurements available are subject to measurement error. System models (consisting of emulator–discrepancy model combinations) are used to estimate storm peak significant wave height (henceforth HS), spectral peak period and storm length jointly in the Danish sector of the North Sea. Using non-stationary extreme value analysis of system output HS, we estimate its 100-year maximum distribution from two different system models, the first based on 37 years of wind–wave simulation, the second on 1200 years; estimates of distributions of 100-year maxima are found to be in good general agreement, but the influence of different sources of uncertainty is nevertheless clear. We also estimate the distribution of 100-year maximum HS using non-stationary extreme value analysis of storm peak wind speed, propagating simulated extreme winds through a system model for HS; we find estimates to be in reasonable agreement with those based on extreme value analysis of HS itself.