Title: Exceedance-based nonlinear regression of residual dependence in extremes
Authors: Linda Mhalla - HEC Montreal (Canada) [presenting]
Valerie Chavez-Demoulin - University of Lausanne (Switzerland)
Thomas Opitz - BioSP, INRA (France)
Abstract: Most environmental processes are spatial by nature. For such spatial processes, different tail behavior types ranging from asymptotic independence to asymptotic dependence are observed. One way of characterizing different tail decay behavior as events become more extreme is via the Pickands dependence function and the angular dependence function, which are appropriate in the presence of extremal dependence and independence, respectively. Motivated by the analysis of the dependence between high background concentrations of nitrogen dioxide measured at different sites in France, we develop a semiparametric framework to estimate both the Pickands and the angular dependence functions based on the excesses and deficits of a min-projection univariate random variable, respectively. Moreover, the effect of a set of predictors on the dependence functions is taken into account based on a generalized additive modeling framework. The methodology is applied to capture the behavior through time and as a function of the distance between sites of the tail dependence between nitrogen dioxide measurements.