Presented by Patrick Raanes on 2-18-2015
A novel approach to account for model noise in the forecast step of the Ensemble Kalman filter (EnKF) is proposed. The core method is based on the approach of the analysis step of ensemble square root filters (ETKF), and consists in right-multiplying the ensemble matrix during the forecast step by a particular ``transform matrix''. Theoretical advantages include respecting linear constraints, minimizing ensemble disarrangement, preserving covariance structures, and modularity. A fundamental problem due to the size of the ensemble subspace is discussed, and possible solutions that complement the core method are suggested and studied. Benchmarks from twin experiments with simple, low order, nonlinear dynamics indicate improved performance over standard approaches to dealing with model noise such as additive, simulated noise and multiplicative inflation.
- kalman filter
- data assimilation