Extremal shot noises, heavy tails and max-stable random fields

TL;DR

This paper studies extremal shot noise models, their properties, and their connection to max-stable fields, especially under heavy-tailed distributions and high point intensities, relevant for modeling spatial extremes.

Contribution

It establishes new limit theorems linking extremal shot noises with max-stable fields under heavy tails and high intensity regimes.

Findings

Limit theorem for extremal shot noise with heavy-tailed distributions

Properties of max-stable random fields derived from shot noise models

Connection to Peak Over Threshold method in extreme value theory

Abstract

We consider the extremal shot noise defined by $$M(y)=\sup\{mh(y-x);(x,m)\in\Phi\},$$ where $\Phi$ is a Poisson point process on $\bbR^d\times (0,+\infty)$ with intensity $\lambda dxG(dm)$ and $h:\bbR^d\to [0,+\infty]$ is a measurable function. Extremal shot noises naturally appear in extreme value theory as a model for spatial extremes and serve as basic models for annual maxima of rainfall or for coverage field in telecommunications. In this work, we examine their properties such as boundedness, regularity and ergodicity. Connections with max-stable random fields are established: we prove a limit theorem when the distribution $G$ is heavy-tailed and the intensity of points $\lambda$ goes to infinity. We use a point process approach strongly connected to the Peak Over Threshold method used in extreme value theory. Properties of the limit max-stable random fields are also investigated.