Estimating Background Error Covariance in Regional 3DVAR of WRF

In order to explore the statistical structure of background error covariance ( B ) and its impact on initial field and three-dimensional variational data assimilation system, based on day by day WRF model forecast data from May to October in 2008, B is estimated using NMC method. The statistical structure of B is analyzed and validated with single ideal experiment, showing that it can reasonably reflect the geostrophic balance relationships and the relationships among multivariable in middle and lower latitudes. The characteristics of B structure are presented as well.A B file provided in three-dimensional variational data assimilation system, it is a generic background error statistics file called CV3- B that can be used for any resolution or area case. The CV3- B is the NCEP background error covariance, which is estimated in grid space by the NMC method. The statistics are estimated with the differences of 24 and 48-hour GFS forecasts with T170 resolution valid at the same time for 357 cases distributed over a period of one year. The major differences between these two kinds of B are the vertical covariance. CV3- B uses the vertical recursive filter to model the vertical covariance but the modified B uses the empirical orthogonal function (EOF) to represent the vertical covariance. In order to compare the difference of the two B in detail, a simulation experiment for June 2009 with two different B are performed to comparatively analyze the initial fields and precipitation distributing of strong rainfall case, and test the simulation effect in a month precipitation. The results show that different B lead to great differences in data assimilation processes. With the same background fields and observation, the minimizing convergence standards are equal to 0.01, the iterative step is about twenty using updated B , while the iterative step is about forty using CV3- B .And using updated B , for most instances it's monotonic decreasing during the iteration except for few iterative steps, and the decrease velocity is more rapid. But using CV3- B the value fluctuates from time to time. From data assimilation process, the efficiency in the iteration is higher using updated B , and the convergence of object function is steadier. The heavy rainfall process triggered by shear on 30 June 2009 in Yunnan Province is selected to analyze the effect of using different B on initial fields and precipitation distribution.Assimilating the sounding data of Tengchong in the west of Yunnan Province, the increment fields of wind vector on 700 hPa is analyzed. It is found that the increment impact spreads to the whole Yunnan Province using CV3- B , which is unreasonable. Using modified B the increment impact is within the adjacent area of the shear, leading to a relatively reasonable result. The precipitation simulation also indicate that using updated B which is consistent with model fields and all kinds of parameter, the TS in moderate rain or over is higher than using CV3- B .The whole simulating effect using updated B is remarkably superior to that using CV3- B . Thus estimating B afresh is important when three-dimensional variational data assimilation system is applied.