Skip to Main content Skip to Navigation

Méthodes probabilistes d'extraction de signaux cachés appliquées à des problèmes de sciences de l'atmosphère

Julien Gazeaux 1
LATMOS - Laboratoire Atmosphères, Milieux, Observations Spatiales
Abstract : This work is devoted to the study of signal extraction applied to atmospheric sciences. The main thread in the different studies presented here is the detection, estimation and characterization of hidden signals. The probabilistic modeling approach has turned out to be well suited to this problematic. For all the problems considered here, the main objective was to respond to the following questions : Which type of information do we expect to find in our data sets ? Are expected hidden signals actually present in the data sets ? How is it possible to detect the time of occurrence of a phenomenon ? How can it be characterized (timing, amplitude ... ) ? If such a signal is detected, what is the uncertainty associated to the detection ? These questions are tackled through three probabilistic hidden signals models. We have focused on non stationary and multivariate models. Along three probabilistic models describing diverse hidden signals (break of variance, pulse-like signals and shift in the mean), we have also developed associated detection methods. The first model is applied to the detection of polar stratospheric clouds in lidar profiles, the second to volcanic eruptions in time series of sulfate and finally the third is applied to detect the date of the onset of the African monsoon in data related to atmospheric dynamics and precipitation. The various methods use a range of techniques in probabilistic modeling, from the likelihood ratio maximization associated with hypothesis testing to the resolution of Kalman filters in a non stationary and non linear framework for the decomposition of multivariate series coupled with the detection of hidden signals. The technical difficulties associated to the extraction of hidden signals are analyzed and the performances of the various algorithms are estimated. The results confirm the interest and the potential of probabilistic methods applied to problems of hidden signals in atmospheric sciences.
Complete list of metadatas
Contributor : Catherine Cardon <>
Submitted on : Wednesday, October 31, 2012 - 1:27:17 PM
Last modification on : Friday, January 10, 2020 - 3:42:30 PM
Long-term archiving on: : Friday, February 1, 2013 - 3:39:21 AM


  • HAL Id : tel-00747464, version 1


Julien Gazeaux. Méthodes probabilistes d'extraction de signaux cachés appliquées à des problèmes de sciences de l'atmosphère. Physique Atmosphérique et Océanique []. Université Pierre et Marie Curie - Paris VI, 2011. Français. ⟨NNT : 2011PA066023⟩. ⟨tel-00747464v1⟩