| 초록 |
Background error covariance (B) is an important part in variational data assimilation (Var), which affects the analyses from Var systems greatly. Because the computation and specification of statistics for B needs great data storage and expensive computations, it's difficult to perform related research. The present paper at first clarifies the importance of B and the necessity of modeling it. Then the principle and computation steps for American NMC method are introduced in detail and used to demonstrate how to apply it in the regional 3D-Var. Thereafter the statistics of background error covariance are computed by using the forecast data set from WRF model. The results indicate that: making use of balance transforms and regression coefficients, control variables are kept to be small values'which ensure the quality of analyses. The first global eigenvector has the maximum component nearby 200hPa, where the error of westerly jet exists. There is negative correlation between low and middle-high levels of first five global eigenvectors of stream function. Compared to lengthscales of stream function and unbalanced velocity potential, those of unbalanced temperature and relative humidity are very small, which show that they are local variables. Lengthscales of stream function and unbalanced velocity potential decrease with the number of vertical mode, while those of unbalanced temoerature and relative humidity vary smoothly. |
