All the computations associated with the operational meteorological da
ta assimilation are usually carried out after the so called cut-off-ti
me, i.e. a few hours after the end of the assimilation period. In this
article we propose an alternative implementation of the four-dimensio
nal variational assimilation where some of the computations occur duri
ng the 24-hour assimilation period. The feasibility of the approach is
first validated with an assimilation system built around a low resolu
tion barotropic grid point model. The results are then confirmed with
a multi-level primitive equation model using real observations. The ma
in result of these experiments is that the peak computer power require
ment of the four-dimensional variational data assimilation may be sign
ificantly reduced by the suggested approach.