Storm peak validation and analysis of uncertainty in estimates of extreme sea states

Cresswell, D. and Via Estrem, L. (2016) Storm peak validation and analysis of uncertainty in estimates of extreme sea states. In: WISE meeting 2016 CNR-ISMAR, 22-26 May 2016, Venice.

PDF TP-036_Storm-peak-validation-and-analysis-of-uncertainty-R2.pdf Download (3MB)

Abstract

Storm peaks are often under-estimated in numerical models, as is widely acknowledged. Yet, for many sites where estimates of extreme storm conditions are needed for engineering design, numerical models are the best – if not only – source of information. We discuss uncertainty in estimates of extreme sea states based on numerical model datasets . A method of validation for extreme conditions is presented based on matched pairs of independent storm peak events between modelled time series and buoy based observations, and found to be preferable to exceedance based validation techniques. Systematic biases in the storm peaks of the CFSR, ERA- Interim and NORA10 datasets, when compared to buoy data, are discussed. Sea states at the peaks of storms are compared and contrasted to the general population of sea states. We demonstrate a calibration scheme designed to remove bias from estimated storm peaks using a minimal set of parameters, in order that parameters may be mapped and estimated at sites where no observations are available. Bias corrected model estimates of storm peaks have a remaining uncertainty associated with model precision. Probability distributions of extreme sea states are estimated using Markov Chain Monte Carlo and Bayesian techniques. Estimates of model precision, based on peak-focussed validation, are used to predict extreme sea states with associated uncertainty representing both model precision and sampling of the long-term distribution. The analysis allows investigation of the relative contribution to estimate uncertainty of model precision and sampling/length of record. Once uncertainty is considered in the estimation of extreme sea states, a contribution to the mean estimate from precision based uncertainty in the source data becomes apparent. This contribution is also relevant to estimates of extreme conditions based on buoy data with measurement and short-term sampling uncertainties, and provides a baseline level of achievable uncertainty in extreme conditions. Validating uncertainty in estimates requires analysis at a large number of sites, and depends to a great extent on the probability associated with the most extreme events in a series, the so-called plotting position. The formulation of the plotting position has been debated for many years. Numerical experiments are presented that suggest that the correct formulation lies within a narrow subset of the debated schemes.

Actions (for site administrators only - login required)

View Item