The initial error growth and predictability in a numerical forecast model are strongly influenced by the observational sample size and distribution and by the method used to project information and error from observations and a first guess (background) field onto the forecast model in the data assimilation process. In order to apply a method of data assimilation, assumptions must be made regarding the structure of error in the first guess fields used to generate the initial conditions. Data assimilation methods are partly characterized in how the three-dimensional error structure is derived and applied.

This study will examine the sensitivity of mesoscale model forecast solutions of the Navy's Coupled Ocean/Atmosphere Prediction System (COAMPS) to several methods of data assimilation; two- and three-dimensional multiquadric interpolation, developed at the Naval Postgraduate School, and optimal interpolation and, if available, three-dimensional variational analysis, developed at the Naval Research Laboratory. Mesoscale model forecast sensitivity to data assimilation methodology will be presented using case studies from cold and warm seasons along the U.S. West Coast.