Data Assimilation with the Local Ensemble Transform Kalman Filter: Addressing Model Errors, Observation Errors and Adaptive Inflation

Our work has addressed several issues relating to Ensemble Kalman Filter (EnKF) for assimilating real data, 1) model errors, 2) inconvenience or infeasibility of manually tuning the inflation factor when it is regional and/or variable dependent and 3) erroneously specified observation error statistics. A Local Ensemble Transform Kalman Filter (LETKF) is used as an efficient representative of other EnKF systems. For the model errors issue, we assimilate observations generated from the NCEP/NCAR reanalysis fields into the SPEEDY model. Several methods to handle model errors including model bias and system-noise are investigated. We address the second and third issues by simultaneously estimating both inflation factor and observation error variance on-line. Our research in this book suggests the need to develop a more advanced LETKF with both bias correction and adaptive estimation of inflation within the system.