Ensemble singular vectors as additive inflation in the Local Transform Kalman Filter (LETKF) framework with a global NWP model

Abstract

We test an ensemble data assimilation system using the 4-D Local Ensemble Transform Kalman Filter (4D-LEKTF) for a global Numerical Weather Prediction (NWP) model with unstructured grids on the cubed sphere. It is challenging to selectively represent structures of dynamically growing errors in background states under system uncertainties such as sampling and model errors. We compute Ensemble Singular Vectors (ESVs) in an attempt to capture fast growing errors on the subspace spanned by ensemble perturbations, and use them as additive inflation to enlarge the covariance in the area where errors are flow-dependently growing. The performance of the 4D-LETKF system with ESVs is evaluated in real data assimilation, as well as Observing System Simulation Experiments (OSSEs). We find that leading ESVs help to capture fast growing errors effectively, especially when model errors are present, and that the use of ESVs as additive inflation significantly improves the performance of the 4D-LETKF.