GSI-NLS4DVar Data Assimilation System: Observation System Simulation Experiments

A nonlinear least-squares four-dimensional variational (NLS-4DVar) data assimilation system based on the gridpoint statistical interpolation (GSI) was developed. The multigrid NLS-4DVar scheme, which not only minimizes large- and small-scale errors in turn, but also does not require an adjoint model due to the use of ensemble strategy, was adopted. Furthermore, an efficient correlation matrix decomposition approach is utilized to modify localization implementation to enhance its calculation efficiency. Meanwhile, with the efficient localization technique, the calculation efficiency has been greatly improved. Operational conventional and satellite observations with the quality control and observation operator in GSI were assimilated. The performance of the system was investigated using the Advanced Research WRF (ARW) model. The observation system simulation experiments were found that the GSI-NLS4DVar system produced more skillful forecasts than the GSI 3DVar system.