### Description

Estimating the physical conditions, including temperature, salinity and current, of an ocean region can be achieved using a computational model based on the equations of geophysical fluid dynamics. Using appropriate resolution in the computation and including critical physical processes, such models can render good state estimates up to a point. But missing physics and inadequate estimates of initial conditions doom the model for useful prediction. This lack of predictive capability can be significantly ameliorated by the systematic assimilation of data into the model so as to correct its output by re-initializing the system state during the calculation and correcting the parameters representing the sub-grid scale and missing processes.

The mathematical procedure of data assimilation (DA) is conceptually well founded on Bayes’ formula of conditional probability. But this is only to say that there is a framework within which to systematically develop methodology for full Bayesian DA in ocean state estimation; prediction remains at a primitive stage of development. There are, of course, an abundance of off-the-shelf techniques, but these are largely based on some linearization procedure implicit in the Kalman-based and variational methods. A fully Bayesian approach that renders uncertainty of the predictions, allows incorporation of diverse data sets and captures the nonlinearity of the underlying system remains a goal with many fundamental as well as practical challenges.

The mathematical procedure of data assimilation (DA) is conceptually well founded on Bayes’ formula of conditional probability. But this is only to say that there is a framework within which to systematically develop methodology for full Bayesian DA in ocean state estimation; prediction remains at a primitive stage of development. There are, of course, an abundance of off-the-shelf techniques, but these are largely based on some linearization procedure implicit in the Kalman-based and variational methods. A fully Bayesian approach that renders uncertainty of the predictions, allows incorporation of diverse data sets and captures the nonlinearity of the underlying system remains a goal with many fundamental as well as practical challenges.

Status | Finished |
---|---|

Effective start/end date | 3/1/15 → 2/28/18 |

### Funding

- DOD DN Office of Naval Research (ONR)

### Fingerprint

data assimilation

ocean

prediction

fluid dynamics

nonlinearity

physics

salinity

methodology

temperature