The NRL Atmospheric Variational Data Assimilation System (NAVDAS) is a modern variational data assimilation system, which is currently used by the Fleet Numerical Meteorology and Oceanography Center (FNMOC) in its operational global and mesoscale numerical weather prediction systems. It is being adapted to provide wind analyses every hour over the Pacific Ocean in support of FAA controlled flight operations. For this role, NAVDAS is run with the On-Scene version of the Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS-OSŪ), which provides GUI level control of model grid parameters and web-based graphics of analyses and forecasts on a variety of computer platforms. In the past year, NAVDAS has been ported to the COAMPS-OS system and extensively tested on small local Linux clusters, a new Linux cluster at FNMOC to be used for operations and a number of DOD (IBM, SGI-ALTIX) supercomputers. The COAMPS-OS framework provides user-friendly interface for relocating grids anywhere over the globe and portability to many computer systems.

In the operational application at FNMOC, NAVDAS is run with COAMPS for a fixed number of operational areas, which are scheduled to run after the boundary conditions are generated by the current run of the global NOGAPS weather forecast model. When running with COAMPS-OS at local and remote sites, a more robust on-demand procedure is needed, where the most up-to date and complete set of NOGAPS boundary conditions that can be obtained at the site are used. To support hourly analyses, the existing aircraft and satellite wind quality control procedures are being adapted to process these observations hourly. In addition, conventional surface and upper air radiosonde observations, aircraft temperature observations, NESDIS satellite temperature retrievals, and SSM/I surface marine winds speeds are utilized. When running NAVDAS with the nested-grid COAMPS model, innovations (observation-background) are obtained by successively interpolating the forecast fields to the observation locations and subtracting the result from the observations for each of the successively finer grid meshes. At the end of the procedure each observation innovation is therefore computed from the highest resolution background forecast available. An analysis grid is constructed by removing any duplicate grid points from the COAMPS grid meshes. Following the analysis, the analysis corrections are added back to each background grid to provide the initial conditions for the next forecast run. An additional correction is added to the outer grid meshes to account for differences between the background forecasts on the grid meshes.