Spatial Bayesian hierarchical modelling of extreme sea states. (November 2016)

• Record Type: Journal Article
• Authors: Clancy, Colm; O'Sullivan, John; Sweeney, Conor; Dias, Frédéric; Parnell, Andrew C.
• Is Part Of: Ocean modelling. Volume 107(2016:Nov.)
• Page Start: 1
• Page End: 13
• Publication Date: 2016-11
• Subjects: Bayesian hierarchical modelling -- Spatial modelling -- Extreme value analysis -- Ocean waves -- Significant wave height; Oceanography -- Periodicals; Océanographie -- Périodiques; Oceanography; Periodicals; 551.46
• DOI: 10.1016/j.ocemod.2016.09.015
• Languages: English
• ISSNs: 1463-5003

Abstract

Highlights: A Bayesian hierarchical framework is used to spatially model extreme sea states. Bayesian inference results in much less uncertainty compared with maximum likelihood. Significant wave height off the west of Ireland can reach 17.5 m with 100-year return period. Abstract: A Bayesian hierarchical framework is used to model extreme sea states, incorporating a latent spatial process to more effectively capture the spatial variation of the extremes. The model is applied to a 34-year hindcast of significant wave height off the west coast of Ireland. The generalised Pareto distribution is fitted to declustered peaks over a threshold given by the 99.8th percentile of the data. Return levels of significant wave height are computed and compared against those from a model based on the commonly-used maximum likelihood inference method. The Bayesian spatial model produces smoother maps of return levels. Furthermore, this approach greatly reduces the uncertainty in the estimates, thus providing information on extremes which is more useful for practical applications.