1986 Volume 64A Pages 61-74
Bratseth (1986) has proposed a successive correction scheme for numerical analysis where the solution converges toward the solution obtained by optimal interpolation. This paper describes how this method is used operationally in two limited area prediction systems, with a mesh of 150 km and 50 km respectively. Data quality control is first performed in a seperate run, and the rate of convergence is used to identify errors. In the following analysis the scale of the horizontal correlation function is decreased with the iteration number. The principle of optimal interpolation is then abandoned, however, the procedure might be looked upon as a simple way of adapting the error statistics to the actual situa. tion. The convergence of the method is investigated, and 5-10 iterations are found to be sufficient for the operational analyses. The method is simple and flexible, and is believed to have advantages compared to traditional optimal interpolation.