This talk has been automatically captioned. You can remove these by pressing CC on the video toolbar.

Name: Christian Rohrbeck

Talk Title: Simulating flood event sets using extremal principal components

Abstract: Catastrophe models are an important tool to estimate the impact of natural hazards. Such models are, for instance, applied by insurance companies to predict the financial capital required to cover potential future payouts. One component of these models is a simulated hazard event set, which represents, for instance, a collection of floods over a long period, e.g., 1,000 or 10,000 years.

In this talk, we introduce a new approach in extreme value analysis to generate hazard event sets in various settings. The methodology is illustrated using a set of 45 river flow gauges in northern England and southern Scotland, an area which regularly experiences severe flood events. We start by utilising the methodology by Cooley and Thibaud (2019) to study extremal dependence across river flow gauges. We then introduce our new framework which uses dimension-reduction techniques, in combination with the concept of extremal principal components introduced by Cooley and Thibaud (2019). Specifically, we approximate the joint extreme value distribution by defining a model which fully captures large-scale spatial structures in the extremes and provides a reasonable fit for the local dynamics. Our approach enables us to simulate large hazard event sets in a few seconds, and we will discuss/highlight the agreement of the characteristics of the observed and simulated extreme events.

Joint work with Dan Cooley (Colorado State University)

This talk is a contributed talk at EVA 2021.