Integration of Novel Mathematical and Statistical Techniques to Support Model-based Water Quality Management

Environmental modelling has been an indispensable tool for addressing water quality problems in freshwater ecosystems. A variety of data-oriented and process-based models have been used to examine the impact of external stressors to ecosystem integrity and to set water quality goals. The main objective of this thesis is to introduce methodological refinements of the current generation of water quality models that can strengthen their foundation as a scientific/predictive tool. One of the central water quality issues revolves around the elucidation of the mechanisAms that shape the composition of phytoplankton assemblages and their capacity to predict the occurrence of harmful algal blooms. My research proposes the introduction of an extra layer of causality into plankton models by connecting algal metabolic processes with species-specific morphological features (cell volume, surface-to-volume ratio, shape). This strategy can effectively delineate model parametric uncertainty, and may be an excellent solution to the identifiability problem of complex over-parameterized model. The motivation of this approach was provided by a meta-analysis of 122 models published in the peer-reviewed literature, which provided ample evidence that our ability to reproduce the observed aggregated and compositional phytoplankton variability is limited, despite the plethora of mathematical expressions aiming to describe different ecophysiological facets of species populations. Hierarchical Bayesian inference offers an appealing ad-hoc strategy to gradually increase model complexity and subsequently enhance their mechanistic underpinning, whenever possible and relevant. The advantages of hierarchical Bayes are illustrated with phosphorus loading models and with the development of risk assessment tools of cyanobacteria dominance. The central message of my dissertation is that the integration of process-based and empirical models offers an appealing methodological prospect, as it allows combining our best mechanistic understanding of ecosystem processes and feedback loops, while remaining within the bounds of data-based parameter estimation and therefore facilitating rigorous error analysis.Ph.D