|
Abstract:
|
The 2017 hurricance season saw three major storms: Harvey, Irma, and Maria cause significant damage to US territories. Current extreme value methods are not well-sutied to aggregated risk due to events like these whose affects would appear separately in the data record. In this work we propose the chi-network for exploring the extremal dependence structure of environmental processes. A chi-network is constructed by connecting pairs where their upper tail dependence coefficient (chi), exceeds a prescribed threshold. We show that the estimated number of connections in a chi network will be biased high and we develop a method to correct the bias of the resulting networks using an empirical Bayesian approach. We illustrate the chi network by analyzing the hurricane season maximum precipitation in Gulf Coast and surrounding area. In addition to the chi network which assesses extremal dependence over an extended period of time, we introduce an annual extremal network to assess connections for a given year. Analysis of the annual extremal network for the Gulf Coast suggests the extremal dependence depends on regional meteorological conditions, for example, sea surface temperature.
|
