Analysed data for numerical prediction can be effectively initialized
by means of a digital filter. Computation time is reduced by using an
optimal filter. The construction of optimal filters involves the
solution of a nonlinear minimization problem using an iterative
procedure. In this paper we describe a simple filter based on the
Dolph-Chebyshev window, which has properties similar to those of an
optimal filter: it is shown to be optimal for an appropriate choice of
parameters. It has an explicit analytical expression and is easily
implemented. Its effectiveness is demonstrated by application to
Richardson's forecast: the initial pressure tendency is reduced from
145 hPa/6 h to -0.9 hPa/6 h. Use of the filter
is not restricted to initialization; it may also be applied as a weak
constraint in four-dimensional data assimilation.