Wednesday, 25 January 2017: 1:45 PM
607 (Washington State Convention Center )
Four-dimensional variational (4D-Var) data assimilation (DA) for oceanic and atmospheric data assimilation is widely considered to be one of the most advanced methods of DA available. However, its use in recent years within the community has been reduced for a number of reasons. First among them is the computational cost of the algorithm on modern supercomputers. This is due to the fact that the 4D-Var method relies on many applications of the adjoint (AD) model and, in some cases (e.g. the representer method), the application of the tangent linear (TL) model as well. Both of these models are typically more expensive than the non-linear forward operator. In this present work, a new method for solving the 4D-Var problem is presented, titled here as the Fast-Track 4D-Var (FT4DVAR). This method aims to reduce the total number of AD and TL model applications during the minimization step while closely approximating the optimal solution given by the full 4D-Var method. This presentation presents an overview of the representer method of 4D-Var as well as the new FT4DVAR methodology. Experiments assimilating real observations of surface and sub-surface temperature and salinity, as well as sea surface height and surface velocity measurements into a Gulf of Mexico Navy Coastal Ocean Model (NCOM) is presented. It is shown here that 96-hr forecasts from an analysis produced by the FT4DVAR method is nearly accurate as that produced from the full NCOM-4DVAR system (in terms of fit to available observations).
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner