With CV3, the control variables are in physical space while with CV5, CV6, and CV7, the control variables are in eigenvector space. The major difference between these two kinds of BE is the vertical covariance; CV3 uses the vertical recursive filter to model the vertical covariance but the others use an empirical orthogonal function (EOF) to represent the vertical covariance. The recursive filters to model the horizontal covariance are also different with these BEs. We have not conducted the systematic comparison of the analyses based on these BEs. However, CV3 (a BE file provided with our WRFDA system) is a global BE and can be used for any regional domain, while CV5, CV6, and CV7 BE’s are domain-dependent, and so should be generated based on forecast or ensemble data from the same domain.