A system identification view on two aquatic topics: phytoplankton dynamics and water mass mixing

The general aim of this study is to modify or improve existing modelling procedures in order to extract more or more reliable information from observations at hand. In practice, this means that much attention is directed to quantifying uncertainties, since these enable to draw a line between what is significant and what should not be interpreted. Two applications were considered: tracer experiments modelled by compartmental models (Part A) and a multivariate water mass mixing model (Part B). However, the focus lies on the methods used and developed to improve the models and their inferences. Indeed, the merit of this work is not to have enabled the estimation of flux rates and mixing fractions, but to have enabled their accurate estimation, together with an estimate of the associated uncertainty. Very briefly, these are the main achievements: (i) Inclusion of the input uncertainties in the estimation of model parameters and their uncertainties.(ii) Model selection method based on the statistical interpretation of the residual Weighted Least Squares cost function.(iii) Improvement of Optimum Multiparameter analysis for large-scale reconstruction of mixing water mass distributions.(iv) Construction of an algorithm to estimate heteroscedastic noise variances, from residuals but corrected for model errors.(v) Verification of identifiability of given experiment-model combinations