From the simple Lorenz-95 model study using small number of ensemble members (N=10), HYB(ENS) and HYB(FD) have the best performance with 0.3 and with 0.6-0.7 weighting factor (a) between the flow-dependent error covariance and the static error covar...
From the simple Lorenz-95 model study using small number of ensemble members (N=10), HYB(ENS) and HYB(FD) have the best performance with 0.3 and with 0.6-0.7 weighting factor (a) between the flow-dependent error covariance and the static error covariance, respectively. However as the number of ensemble members increases, HYB(ENS) shows the best results with a=0.1. Also, EnKF and HYB(ENS) have almost the same performance with larger ensemble members (N=20). Consequently, HYB(FD) shows that the errors are decreased by about 20 % compared to pure 3dvar on the simple model test.