A Bayesian spatial hierarchical model for extreme precipitation in Great Britain

TL;DR

This paper introduces a Bayesian hierarchical spatial model for extreme precipitation in Great Britain, improving the spatial consistency and precision of risk estimates for flooding-related events.

Contribution

It develops a novel Bayesian hierarchical approach that incorporates spatial dependence into extreme value analysis, enhancing the accuracy of precipitation risk estimates.

Findings

Improved spatial consistency of return level estimates.

Enhanced precision of extreme precipitation risk estimates.

Effective modeling of spatial dependence in precipitation data.

Abstract

Intense precipitation events are commonly known to be associated with an increased risk of flooding. As a result of the societal and infrastructural risks linked with flooding, extremes of precipitation require careful modelling. Extreme value analysis is typically used to model large precipitation events, though a site-by-site analysis tends to produce spatially inconsistent risk estimates. In reality, one would expect neighbouring locations to have more similar risk estimates than locations separated by large distances. We present an approach, in the Bayesian hierarchical modelling framework, that imposes a spatial structure on the parameters of a generalised Pareto distribution. In particular, we look at the clear benefits of this approach in improving spatial consistency of return level estimates and increasing precision of these estimates. Unlike many previous approaches that pool information over locations, we account for the spatial dependence of the data in our confidence intervals. We implement this model for gridded precipitation data over Great Britain.