The requirement to modify meteorological analyses to avoid spurious high frequency oscillations in numerical forecasts has been known from the beginning of numerical weather prediction. The most popular method of initialization up to recently was normal mode initialization, or NMI (Machenhauer, 1977). This has been used in many NWP centres, and has performed satisfactorily. Its most natural context is for global models, for which the horizontal structure of the normal modes corresponds to the Hough functions, the eigenmodes of the Laplace Tidal Equations. For limited area models, normal modes can also be derived, but the lateral boundaries force the introduction of simplifying assumptions.
Recently, an alternative method of initialization, called digital filter initialization (DFI), was introduced by Lynch and Huang (1992). It was generalised to allow for diabatic effects by Huang and Lynch (1993). The latter paper also discussed the use of an optimal filter. A much simpler filter, the Dolph-Chebyshev filter, which is a special case of the optimal filter, was applied to the initialization problem by Lynch (1997). A more efficient formulation of DFI was presented by Lynch, Giard and Ivanovici (1997).