Abstract
The use of dynamic initialization is examined first for the case of a linearized barotropic model in which both the meteorological and gravity‐wave modes have non‐zero frequencies, and all variables are allowed to adjust mutually. It is shown that the gravity‐wave mode is rapidly damped, leaving the meteorological mode almost intact with appropriate values of divergence as well as balanced mass and vorticity fields. There are some problems associated with the slower rate of convergence at large scales; it is suggested how these might be overcome using spectral rather than finite‐difference methods. Mutual adjustment of the variables is shown to be generally more appropriate than forced adjustment.
The arguments are extended to a multi‐level model by diagonalizing it into its vertical normal modes. The analysis indicates that the external gravity‐wave mode should be rapidly damped by dynamic initialization, giving appropriate values of the vertically‐meaned divergence, while low‐frequency internal modes will be relatively unaffected by the initialization procedure.
A series of experiments is performed using a 5‐level hemispheric model, and the predictions of the linearized theory are confirmed. In particular it is found that the amount of computation required to reduce the amplitude of the external gravity‐wave mode to small levels is equivalent to that required for only 2–3 hours of forward integration. However, the quality of the forecast appears to be surprisingly insensitive to the presence or absence of gravity waves, and the need for such an initialization procedure is not conclusively demonstrated.