Abstract


TOWARDS A BETTER UNDERSTANDING OF PHYTOPLANKTON BLOOM DYNAMICS AND ASSESSMENT OF ECOLOGICAL STATUS USING UNSUPERVISED DYNAMIC MODELLING

Despite recent scientific improvements, it is obvious to further knowledge on phytoplankton bloom dynamics for scientific issues and also to help stakeholders to meet requirements of EU Directives. Consequently, there is a rising trend in implementing high spatial and temporal sampling strategy involving systems generating large amounts of data. A good knowledge of these data is a hard task owing to the important variability in ecosystems dynamics and nature of the data. A dynamic model based on an Unsupervised Hidden Markov Model (UHMM) is developed. A spectral clustering is used to define UHMM structure without assumptions about data distribution and so, environmental states. States are characterized by a specific combination of all the input measurements and its temporal dynamics. The UHMM with Viterbi algorithm allows to develop early warning system for phytoplankton blooms and near-real time classification of each new incoming data set in a given environmental states. Two databases, from a high resolution instrumented station (MAREL Carnot, Ifremer), and from a pocket Ferry-Box implemented in the eastern English Channel, are used to highlight the added-value of the proposed approach.

Authors

ROUSSEEUW, K., IFREMER, France, kevin.rousseeuw@ifremer.fr

POISSON-CAILLAULT, E., ULCO, France, poisson@lisic.univ-littoral.fr

LEFEBVRE, A., IFREMER, France, alain.lefebvre@ifremer.fr

Details

Oral presentation

Session #:085
Date: 2/24/2015
Time: 16:00
Location: Machuca (Floor -2)

Presentation is given by student: No