Four-Dimensional Variational Assimilation of Precipitation Data

Abstract

The benefits of assimilation of precipitation data had been demonstrated in diabetic initialization and nudging-type experiments some years ago. In four-dimensional variational (4DVAR) data assimilation, however, the precipitation data have not yet been used. To correctly assimilate the precipitation data by the 4DVAR technique, the problems related to the first-order discontinuities in the “full-physics” forecast model should be solved first. To address this problem in the full-physics regional NMC eta forecast model, a modified, more continuous version of the Beta-Miller cumulus convection scheme is defined and examined as a possible solution. The 4DVAR data assimilation experiments ate performed using the conventional data (in this case, analyses of T, ps, u, v, and q) and the precipitation data (the analysis of 24-h accumulated precipitation). The full-physics NMC eta model and the adjoint model with convective processes are used in the experiments. The control variable of the minimization problem is defined to include the initial conditions and model's systematic error parameter. An extreme synoptic situation from June 1993, with strong effects of precipitation over the United States is chosen for the experiments. The results of the 4DVAR experiments show convergence of the minimization process within 10 iterations and an improvement of the precipitation forecast, during and after the data assimilation period, when using the modified cumulus convection scheme and the precipitation data. In particular, the 4DVAR method outperforms the optimal interpolation method by improving the precipitation forecast.