A full-physics adjoint model is used as a bench model to examine the impact of various physical parameterizations on four-dimensional variational data assimilation. Based on the understanding of the impact of physical parameterizations on 4D Var, a new minimization algorithm is then formulated. This algoritm takes into account the impact of physical processes but incurs only a minimal increase in computational cost. It is a generalization of the truncated Newton method, and is characterized by use of different simplifications of the physical parameterization package at different stages of the minimization process.

Key words: variational data assimilation, full-physics adjoint, minimization algorithm