 |
 |
School of Physical, Environmental and Mathematical Sciences |
 |
|
Field of Study - Oceanography
Supervisor: Dr Andrew Kiss (PEMS)
Co-Supervisor: Dr Oscar Alves (Bureau of Meteorology)
Research Topic - Ensemble Methods in Oceanic Modelling
Brief Summary of Work
This research project seeks to investigate the predictability of oceanic models, with particular emphasis on the Limited Area Model (LAM) which is a component of Project BLUElink (www.bom.gov.au/bluelink/). This LAM is the CSIRO Marine and Atmospheric Research (CMAR) division's 'Sparse Hydrodynamic Ocean Code' (SHOC).
A central challenge of ensemble techniques is that the number of degrees of freedom of the model under investigation usually exceeds the number of ensemble members that can practicably be studied, often by several orders of magnitude. Although this challenge can be mitigated to some extent by constraining the domain size and resolution of the model, additional strategies are also required to prevent the computational burden from becoming excessive.
Two main approaches have so far been pursued. Firstly, SHOC has been set up over a 175 km x 110 km area off the coast of Perth, and error growth has been studied by perturbing sea surface temperature (SST), using Empirical Orthogonal Functions (EOFs). This enables the Jacobian matrix to be approximated with a relatively small number of model runs. Secondly, a 'toy model' has been set up over a small sub-domain of the Perth domain, approximately 20 km x 20 km in size. The toy model is small enough that the entire model state vector has only 1,229 elements, so the whole of the model phase space can therefore be investigated. It is intended that the toy model will be used to develop and confirm the applicability of perturbation techniques, which will then be applied to the larger domain.
SHOC is hosted on a compute cluster in the CSIRO / Bureau of Meteorology (BoM) High Performance Computing and Communications Centre (HPCCC) (http://www.hpccc.gov.au/) at BoM Head Office in Melbourne, and is accessed remotely. Model data is in NetCDF format, and is visualised and manipulated using MatLab.
The study is being pursued on a part-time basis, under the "Postgraduate Study at ADFA" Scheme. The assistance of CMAR and BoM colleagues, particularly from the HPCCC and those involved in Project BLUElink, is gratefully acknowledged.
Preliminary results were presented in a poster (92 KB) at the 14th AMOS Conference in Adelaide.
An overview of the theoretical background to the project was recently published in an article in the
June 2007 edition (Vol. 20, pp58-63) of the Bulletin of the Australian Meteorological and Oceanographic Society.