Study Of The Statistical Structure Of Background Errors And Its Impact On 3D-Var

Background error covariance is very important to assure exclusive result, to govern the amount of smoothing of the observed information and to decide relationships between different variables in variational data assimilation. To a large extent, the form of this background error covariance governs the resulting objective analysis.Because the research of variational data assimilation in our country is not advanced, there is little result in research of background error covariance, which is becoming a desiderating problem in present numerical weather prediction (NWP) operation and research work. The statistical structure of background error covariance and its impact on three-dimension variational data assimilation system are studied in this paper.In order to get the height-height background error covariance, the innovation vector method is used in this paper. The data consisted of innovation data (12 h and 24 h predicted height of T213 model minus radiosonde measurements) at times 00 UTC and 12 UTC. Horizontal characteristic length, prediction error variance and observation error variance are obtained using Gauss correlation function in a particular level. The straightforward way and the empirical thickness method are used to get approximate function in interlevel values. In vertical direction, vertical covariance approximation is obtained by the second-order autoregressive (SOAR) correlation function and distance transformation method. The resulting three-dimensional approximation function is partially separable, being the product of the horizontal covariance function and the vertical correlation function. Different correlation functions fitting background error covariance are investigated and the results are given.With new statistic background errors and horizontal characteristic length, the impact of background error covariance has been checked by comparing and analyzing the result of assimilation. With the NWP initial field attained by the assimilation, the process of typhoon has been forecasted through the GRAPES model. It was found that new statistic background errors and horizontal characteristic length have the potential to improve the quality of analysis field and the accuracy of track and precipitation forecast of tropical cyclone.