Four Dimensional Variational Data Assimilation for COAMPS

A four dimensional variational data assimilation system has been developed for the US Navy's relocatable limited-area atmospheric model. The adjoint and tangent linear COAMPS (Coupled Ocean Atmosphere Mesoscale Prediction System) models propagate observational information backward and forward in time during the iterative minimization process. The background error covariances of the state variables are modeled with a diffusion operator. When compared to global systems, smaller effective horizontal grid spacing and lateral boundary conditions are challenging design considerations. The model's physical domain can be placed anywhere on Earth, so the data assimilation process must be robust for all locations. The system's performance in areas of varying topography and flow regimes (tropic, mid-latititudes, etc.) is satisfactory when compared to the existing three dimensional variational system. The system is also designed to account for model error through a weak constraint formulation. This feature will become more important as the model's resolution and complexity increases.