Seminars

Seminar Abstract

Representing model error in an Ensemble Kalman Filter:
application to an eddy-resolving ocean model

The Ensemble Kalman filter (EnKF) is a data assimilation method that uses an ensemble of model runs to estimate the model and forecast error in order to implement the Kalman Filter equations. A serious limitation of the EnKF is the progressive collapse of the ensemble after each assimilation cycle. In order to prevent this collapse, we seek to develop a method for representing model error in an Ensemble Kalman Filter for application to an eddy-resolving ocean circulation model. Several methods have been proposed in the literature: for example treating the observations as random variables, applying an inflation factor to artificially maintain ensemble spread, sub-sampling of a much larger matrix in order to increase the rank of the ensemble and the introduction of spatially correlated noise.

There are several elements of an ocean model that are known to introduce errors - for example mixing and advection schemes. By performing a series of model runs using several different schemes, we attempt to diagnose the magnitude of the model error that this uncertainty introduces. We analyse the spread of the terms in the model equations to derive an appropriate scaling of these terms. From this analysis, we introduce a random, time-correlated, rescaling of the uncertain terms in the model equations in order to represent model error in each ensemble member.

[Back to Seminars]

CSIRO = Marine Laboratories Auditorium, Castray Esplanade, Hobart

For further information, or to schedule a seminar, contact:
Karen Wild-Allen, (Oceanographic seminars) CSIRO Marine and Atmospheric Research (03) 6232 5010
Piers Dunstan, (Biological seminars) CSIRO Marine and Atmospheric Research (03) 6232 5382
Sandra Zicus, Antarctic Climate and Ecosystems CRC (03) 6226 7888 & Institute of Antarctic and Southern Ocean Studies (IASOS) University of Tasmania (03) 6226 2509