Winter/Spring 2011

Marine hydrodynamic and biogeochemical models are essential tools for the assessment and prediction of issues ranging from the effects of global climate change through to the response of bays and estuaries to point-source nutrient loads. There has been a substantial effort to quantify and explore hydrodynamic and biogeochemical model error, using a variety of techniques under the general headings of model-data fusion or data assimilation. We focus on two distinctly different approaches to reduce and/or quantifying model error:

1.) An Ensemble Optimal Interpolation (EnOI) data assimilation scheme is applied to a coastal hydrodynamical model of south-east Tasmania, Australia, and;

2.) Bayesian Hierarchical Modeling in a 0-D BGC mixed-layer model of Ocean Station Papa in the north-east Pacific Ocean.

The EnOI system used in the hydrodynamic model corrects for errors in the forcing data and initial conditions and uses observational data from a mooring array and Slocum Gliders. The EnOI algorithm, in its current form, is unable to give any estimates of the uncertainty in the model parameters and suffers from a number of underlying assumptions. On the other hand, Bayesian Hierarchical Modeling (BHM) provides a rigorous probabilistic framework to account for uncertainty in parameters, state variables, and initial conditions in the presence of .noisy. and incomplete data. The BHM framework can incorporate prior knowledge about uncertain quantities, using probability distributions, for model parameters and/or the initial conditions for state variables. Bayesian inference yields a joint posterior probability distribution for the model state and parameters given the data. However, our current inference engine is computationally very expensive and presently intractable in 3D problems. To demonstrate the BHM framework we have applied it to a simple 0D mixed layer BGC model (NPZD).Using twin-experiments and the historical Ocean Station P dataset we investigate what we can really learn about the state and model parameters for BGC systems,under different observation scenarios.

BIO: "Emlyn Jones is working as a post-doc at the Commonwealth Scientific and Industrial Research Organisation (CSIRO) located in Hobart, Australia. He completed his PhD at Flinders University, South Australia in 2009. His background is in numerical modelling (waves and hydrodynamics) and sediment-pelagic interactions. Before working on new model-data fusion techniques he admits to often tuning and calibrating deterministic models in an ad-hoc manner, but now realises that perhaps there was/is a better way ... "

Gary will discuss problems that arise when trying to predict the ocean mesoscale. The talk will use a combination of observations, complex and idealized ocean models, and ocean dynamics. The seminar will be informal with lots of opportunities to ask questions.