In this study, retrospective optimal interpolation (ROI) technique is examined in the WRF simulation. ROI iteratively assimilates an observation set at a post analysis time into a prior analysis at the analysis time. The final analysis of ROI is the same as the global minimum of 4D-Var if the relaxation is successful. By exploiting the perturbation method, we can implement ROI without using an adjoint model. Considering the effective computation, to decrease the rank of the error covariance is needed when the ROI method is used.

The ensemble using 30-perturbed datasets which are produced by adding random perturbation with standard deviations of 3 m/s and 3 K to the horizontal wind and temperature at the initial time is analyzed to find the adaptive regions and variables. To decrease the rank of the error covariance, we chose the sensitive regions and variables from the ensemble results which are performed in the WRF model. Using the results which are chosen, we made the several matrix of the error covariance with various ranks. Finally, the effects of the data assimilation using ROI are verified in this study.