at the Endoume Marine Station (Building 4, Conference Room, Chemin de la Batterie des Lions, 13007 Marseille)
On the following topic: "Neural methods and mixed data: Towards better spatio-temporal resolution of marine ecosystems and phytoplankton"
Summary:
Phytoplankton is one of the first links in the food web and generates
up to 50% of the world's primary production. The study of phytoplankton and its
The physical environment requires observations with a resolution of less than
per day and per kilometre, as well as taking into account heterogeneous types of data
involved and the spatio-temporal dependency structures of ecosystems
sailors.
This thesis aims to develop statistical methods in this context by
using technologies such as automated flow cytometry. The developments
Theoretical work has focused on deep Gaussian mixture models
(DGMM) introduced by Viroli and McLachlan (2019). In order to better characterise
ecological niches of phytoplankton, we have extended these models to data from
mixed (with continuous and non-continuous variables) often present in
A clustering method has been proposed, as well as an algorithm
for generating synthetic mixed data.
With regard to the high-frequency study itself, neural networks
have been introduced to process flow cytometry output and study
six functional groups of phytoplankton in coastal zones and open oceans. From
differentiated and reproducible reactions of these groups have been identified following
wind-induced impulse events, underlining the importance of coupling
between physics and biology. In this respect, a break detection method has been developed.
has been proposed to delimit the epipelagic and mesopelagic zones, thus proposing
a new basis for calculating carbon mesopelagic budgets.
Keywords:Mixing models, Synthetic data, Break detection, Phytoplankton,
Ecological niches, Physical forcing
Composition of the jury
Denys POMMERET Professor Aix Marseille University - Institut de Mathématiques de Marseille (I2M)
163 Avenue de Luminy Campus de Luminy 13009
Marseille FRANCE Thesis supervisor
Christophe BIERNACKI Professor University of Lille 1 - Paul Painlevé Laboratory
Building M2 Cité Scientifique 59655
Villeneuve-d'Ascq FRANCE Rapporteur
Emilie POISSON-CAILLAULT Associate Professor Université du Littoral Côte d'Opale (ULCO) - Signal and Image Computing Laboratory of the Côte d'Opale (LISIC)
Maison de la Recherche Blaise Pascal BP 719, 50, rue Ferdinand Buisson 62228
Calais FRANCE Rapporteur
Melilotus THYSSEN Research Fellow Aix Marseille University - Mediterranean Institute of Oceanology (MIO)
Bâtiment Océanomed - Méditerranée 163 Avenue de Luminy Campus de Luminy 13009
Marseille FRANCE Thesis co-director
Melika BAKLOUTI Professor Aix Marseille University - Mediterranean Institute of Oceanology (MIO)
Bâtiment Océanomed - Méditerranée 163 Avenue de Luminy Campus de Luminy 13009
Marseille FRANCE Examiner
Pierre LATOUCHE Professor University of Paris - MAP5 Laboratory, Ecole Polytechnique
45 rue des Saints-Pères Laboratoire MAP5 Aile Turing 75006
Paris FRANCE Examiner
Geoffrey MCLACHLAN Professor University of Queensland - School of Mathematics and Physics
University of Queensland St Lucia 4067
Brisbane AUSTRALIA Examiner
Mridul K. THOMAS Maître assistant University of Geneva - Dept. F.-A. Forel for Environmental and Aquatic Sciences
A209, Uni Carl Vogt 66, Boulevard Carl-Vogt 1211
Geneva SWITZERLAND Examiner