A0744
Title: Spatial clustering of multivariate time series based on extremal dependence between sites
Authors: Gwladys Toulemonde - University of Montpellier; Inria (France) [presenting]
Alexis Boulin - Universite Cote d'Azur and Inria, Lemon (France)
Elena Di Bernardino - University Cote de Azur (France)
Thomas Laloe - LJAD - Universite de Nice (France)
Abstract: Disastrous climate events such as floods or wildfires often occur due to the simultaneous extreme behavior of several interacting processes. The main objective is to develop high-dimensional clustering techniques to handle compound extreme events. The first step consists of developing an extremal dependence summary measure between random vectors in order to quantify extremal dependence between sites on which not only one but several time series could be then considered. The proposed measure becomes a key ingredient in proposing a clustering algorithm for multivariate time series. This method is illustrated by proposing a regionalization task based on gridded data from climate models over Europe. More precisely, based on the ERA5 reanalysis dataset from 1979 to 2022, both daily precipitation sums and daily maximum wind speed data are considered in each pixel.
