Diagnostics for pairwise extremal dependence in spatial processes (Q1424684): Difference between revisions

This paper deals with diagnostic procedures for identifying the main structure of the dependence of extreme spatial events. The authors consider the class of spatial processes \(\{X(\nu) \}_{\nu\in V}\) in the region \(V\subset R^2\) with marginal distribution \(F_{X(\nu)}(\cdot)\) and such that the marginally standardized process \(F_{X(\nu)}\{X(\nu)\}\) is stationary. Let us consider the stationary spatial process \(\{Z(\nu)\}_{\nu\in V}\), where \(Z(\nu)=-1/\log F_{X(\nu)} \{X(\nu)\}\). For any location \(\nu_{1}, \nu_{2}\in V\), separated by the vector \(\tau=\nu_{2}-\nu_{1}\), because of the stationarity of the process the joint distribution function is \[ \text{Pr}(Z(\nu_{1}) \leq x,\quad Z(\nu_{2}\leq y))= \text{Pr}(Z(\nu)\leq x,\;Z(\nu+ \tau)\leq y)=F_{\tau}(x,y), \] for any \(\nu\in V\) such that \(\nu +\tau\in V\), and the corresponding joint survivor function is denoted by \(\overline F_{\tau}(x,y)\). The joint tail along the diagonal for the process \(\{Z(\nu)\}\) for a fixed \(\tau\) is \(\overline F_{\tau}(z,z)={\mathcal L}_{\tau}(z)z ^{-1/\eta_ {\tau}}\) as \(z\to\infty\), where \(\eta_{\tau}\) termed, the coefficient of tail dependence, is a constant in \(z\) describing the decay rate of the probability of simultaneous exceedances of \(z\), and \({\mathcal L}_{\tau}(\cdot)\) is a slowly varying function. There are presented two measures of pairwise extremal dependence of the spatial process: primary \(\gamma (\tau)=2(1-\eta_{\tau})\), where \(\gamma(\tau)\in [0,2)\) and \(\gamma(\tau)=\gamma(-\tau)\); and, secondary, the measure of pairwise extremal dependence \({\mathcal L}_{\tau}(t)\). The authors illustrate the properties and use of these measures through theoretical examples and applications in hydrology and oceanography.