|
Applied Math Seminar
|
|
To accurately predict future ocean states, it behooves us to use all available observational data. Much of the sub-surface data comes from Lagrangian instruments that are assumed to follow fluid particle trajectories. Model and observation errors are taken to be random processes and filtering methods are used to incorporate data into a re-initialization of the state. Special considerations arise for Lagrangian data as it is not given directly in terms of the state variables. A method is proposed for dealing with this issue and dynamical systems ideas are then invoked to shed light on both the failure of certain filters and the most useful data. |