Abstract:
Based on the GRAPES 3-D variational data assimilation system (p3DVAR), which is defined on isobaric surfaces, a new system (m3DVAR) is reconstructed to match with the nonhydrostatic GRAPES model in order to reduce the errors caused by spatial interpolation and variables transformations, thus improving the quality of initial value for the operational weather prediction model. The analysis variables of the m3DVAR system are completely consistent with the predictands of the GRAPES model in terms of spatial staggering and physical definition. Because of vertical hybrid ordinate and nonhydrostatic conditions, a new scheme for solving dynamical constraint equations is designed for the m3DVAR data assimilation system. To deal with the difficulties in solving the nonlinear balance equation at sigma levels, dynamical balance constraints between mass and wind fields are reformulated, and an effective mathematical scheme is implemented to avoid complicated problems caused by the nonlinear equations under the terrain following coordinates. Meanwhile, new observation operators are designed for routine observational data, and background error covariance is also calculated. Currently, the m3DVAR system is able to assimilate all the conventional observations data, such as sounding, surface and ship data, etc. The multivariate ideal experiments with single point observation are performed to test validity of the system. The results show that the m3DVAR data assimilation system is able to correctly describe the multivariate analysis and the relationship of physical constraints. The differences of innovation and analysis residual for π also show that the analysis error of the m3DVAR system is less than that of the p3DVAR system. The Ts scores of the 24-hour precipitation prediction in August 2006 indicate that the m3DVAR system has smaller errors of model initial values relative to the p3DVAR system. So, the m3DVAR system can improve the analysis quality and initial values for the GRAPES model.