A Bayesian approach to state and parameter estimation in a Phytoplankton-Zooplankton model

Within the last 15 years there have been significant advances in statistical techniques available for data assimilation (DA), largely driven by a combination of the availability of high performance computing and dense observational data-sets (Lorenc 2003; Evensen 2007). Marine Biogeochemical (BGC) modelling, unlike the disciplines of numerical weather prediction and operational oceanography, has been slow in adopting these new and evolving methods. This is not surprising given the lack of high quality data-sets with adequate temporal and spatial resolution. Recent advances in sensor platforms (e.g. coastal gliders, nutrient sensors, etc.) and remote sensing algorithms are providing new data streams that could be used in data assimilation routines for biogeochemical state and parameter estimation. Unlike atmospheric and hydrodynamic models, which are based on sound physical laws such as the Navier-Stokes equations, biogeochemical models represent processes which are highly parametrised and often based on empirical studies (Miller 2004). It is becoming more apparent that the traditional form of the deterministic biogeochemical model using fixed parameters is an unrealistic representation of many key processes included in the current models. The natural diversity of phytoplankton and zooplankton assemblages is represented by a small number (often 2-3) of functional groups, and the variability of community composition within functional groups is ignored. This variability can be overcome by reformulating many of the deterministic parametrisations with stochastic parameters. Given the high prior uncertainty in many biogeochemical parameters, a combination of DA and parameter estimation routines must be invoked to adequately constrain the stochastic parameters. A Bayesian approach to state and parameter estimation in a Phytoplankton-Zooplankton model