Statistics of extreme ocean environments: Non-stationary inference for directionality and other covariate effects

TL;DR

This paper compares various non-stationary extreme value inference methods for ocean environment data, focusing on covariate effects and model parameterisations, to identify the most accurate and efficient approaches.

Contribution

It critically evaluates multiple inference procedures, including different parameterisations and Bayesian versus maximum likelihood methods, for modeling non-stationary extreme ocean data.

Findings

Spline and Gaussian Process models perform best in inference quality and efficiency.

MCMC with mMALA algorithm yields comparable results for spline and Gaussian Process models.

Alternative methods generally underperform compared to spline and Gaussian Process approaches.

Abstract

Numerous approaches are proposed in the literature for non-stationarity marginal extreme value inference, including different model parameterisations with respect to covariate, and different inference schemes. The objective of this article is to compare some of these procedures critically. We generate sample realisations from generalised Pareto distributions, the parameters of which are smooth functions of a single smooth periodic covariate, specified to reflect the characteristics of actual samples from the tail of the distribution of significant wave height with direction, considered in the literature in the recent past. We estimate extreme values models (a) using Constant, Fourier, B-spline and Gaussian Process parameterisations for the functional forms of generalised Pareto shape and (adjusted) scale with respect to covariate and (b) maximum likelihood and Bayesian inference procedures. We evaluate the relative quality of inferences by estimating return value distributions for the response corresponding to a time period of $10 \times$ the (assumed) period of the original sample, and compare estimated return values distributions with the truth using Kullback-Leibler, Cramer-von Mises and Kolmogorov-Smirnov statistics. We find that Spline and Gaussian Process parameterisations estimated by Markov chain Monte Carlo inference using the mMALA algorithm, perform equally well in terms of quality of inference and computational efficiency, and generally perform better than alternatives in those respects.