Detecting the stationarity of spatial dependence structure using spectral clustering - Archive ouverte HAL Accéder directement au contenu
Pré-Publication, Document De Travail (Preprint/Prepublication) Année : 2023

Detecting the stationarity of spatial dependence structure using spectral clustering

Résumé

Modeling extreme events require some knowledge on the spatial stationary of dependence structures in order to construct reliable statistical models. For spatial processes, assuming stationarity of the dependence structure may not be reasonable due to topology of the region under study for example. In this study, we present an adapted spectral clustering algorithm for spatial extremes by considering the extremal concurrence probability as a similarity metric of the dependence structure among the stations. This algorithm involves a heuristic method able to determine whether the dependence structure of the spatial process is stationary or not. It is furthermore able to detect the number of clusters k with high accuracy. In non stationary dependence structure case, the algorithm clusters the stations into k regional clusters with similar dependence structure. In order to validate our proposed methodology, we tested it on different simulation cases based on one or more max-stable models. The accuracy of the results encouraged us to apply it on two real data set: rainfall data in the east coast of Australia and rainfall over France.
Fichier principal
Vignette du fichier
Detecting_the_stationarity(submitted version).pdf (1.41 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03918937 , version 1 (02-01-2023)
hal-03918937 , version 2 (25-04-2023)
hal-03918937 , version 3 (19-02-2024)

Identifiants

  • HAL Id : hal-03918937 , version 1

Citer

Véronique Maume-Deschamps, Pierre Ribereau, Manal Zeidan. Detecting the stationarity of spatial dependence structure using spectral clustering. 2023. ⟨hal-03918937v1⟩
139 Consultations
125 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More