B1961
Title: Modelling non-stationarity in asymptotically independent extremes
Authors: Callum Murphy-Barltrop - Lancaster University (United Kingdom) [presenting]
Abstract: In many practical environmental applications, it is important to evaluate the joint extremal risk from two or more variables. However, the variables of interest often exhibit non-stationarity: consequently, the vast majority of approaches for multivariate extremes, where data is assumed to be identically distributed, are not applicable in this setting. Moreover, non-stationary trends often exist within marginal distributions and dependence structures simultaneously, resulting in complex data structures. Few approaches have been proposed for capturing such structures in the extremes literature to date. We propose a flexible semi-parametric modelling framework for capturing trends in asymptotically independent extremes. We show this framework is able to accurately capture a broad range of extremal dependence trends across simulate examples. We also demonstrate our approach using temperature data from the UK Climate Projections from 1980-2080. Marginal trends are first accounted for via a pre-processing technique. Our model is then applied to estimate trends in the extremal dependence that are in good agreement with empirical evidence. Finally, the fitted model is used to estimate joint extreme events up to the year 2080, allowing us to analyse of the impact of climate change on such events.
