Détection et identification des types de rupture dans les séries d'extrêmes de précipitation simulés par un modèle régional du climat
Directeur de thèse : Jocelyn CHANUSSOT
École doctorale : Electronique, electrotechnique, automatique, traitement du signal (EEATS)
Spécialité : Signal, image, parole, télécoms
Structure de rattachement : Université Grenoble Alpes
Établissement d'origine : Federal university of Santa Catarina - Brésil
Financement(s) : Bourse erasmus mundus
Date d'entrée en thèse : 01/09/2010
Date de soutenance : 08/10/2013
Composition du jury :
M. Patrick Flandrin, Laboratoire de Physique, ENS Lyon, Président
M. José Bermudez, LPDS, Dept. of Electrical Engineering, Federal University of Santa Catarina (UFSC), Brazil, Rapporteur
M. Cédric Richard, Laboratoire Lagrande, UMR CNRS 7293, Observatoire de la Côte d''Azur, Université de Nice Sophia-Antipolis
M. Pierre Borgnat, Laboratoire de Physique, ENS Lyon, Examinateur
Mme Marie Chabert, INP-ESEEIHT, Toulouse, Examinatrice
M. Philippe Naveau, LSCE, CNRS, Examinateur
M. Jocelyn Chanussot, Gipsa-lab, Grenoble INP, Directeur de thèse
Mme Anne-Catherine Favre, LTHE, Grenoble INP, Co-directrice de thèse
Résumé : Over the last decade, different stationarity tests have been proposed. For testing real world signals, a special attention should be given to the nonparametric techniques, and to the fact that we often do not know wehther a change has occurred nor do we have any idea where the possible change point(s) could be. In this Thesis, different nonparametric techniques are developped fortesting the stationarity and estimating the change point of real workd data, Firstly, a number of contributions are proposed to an existing stationarity test developed in time-frequency domain. Having identified the limitations of the existing techniques, a novel nonparametric stationarity test is then proposed, which is more sensitive to slowly-varying, first-order nonstationarities. After presenting the contributyions to the stationarity tests, an alternative framework that allows for the detection of multiple change points is proposed. Finality, all the methods are applied to an environmental data (rainfall time series), and the sonsistency of the results confirms the potential of the proposed techniques in comparison to other methods in the literature.
