5.1. LETKF AlgorithmΒΆ
The following is a brief conceptual overview from [Sluka2016] of how the LETKF algorithm operates, for a complete description see [Hunt2007].
The local ensemble transform Kalman filter (LETKF) is a type of ensemble Kalman filter (EnKF) which uses an ensemble of forecasts \(\left\{\mathbf{x}^{b(i)} : i = 1,2,...,k \right\}\) to determine the statistics of the background error covariance. This information is combined with new observations, \(\mathbf{y}^o\), to generate an analysis mean, \(\bar{\mathbf{x}}^a\), and a set of new ensemble members, \(\mathbf{x}^{a(i)}\). First, the model state is mapped to observation space by applying a nonlinear observation operator \(H\) to each background ensemble member
note, that the application of the observation operator is applied outside this UMD-LETKF library.
A set of intermediate weights, \(\bar{\mathbf{w}}^{a}\) are calculated to find the analysis mean \(\bar{\mathbf{x}}^a\)
where \(\bar{\mathbf{x}}^b\) and \(\bar{\mathbf{y}}^b\) are the ensemble mean of the background in model space and observation space, respectively. \(\mathbf{X}^b\) and \(\mathbf{Y}^b\) are the matrices whose columns represent the ensemble perturbations from those means, and \(\mathbf{R}\) is the observation error covariance matrix.
Last, the set of intermediate weights, \(\mathbf{W}^a\) are calculated to find the perturbations in model space for the analysis ensemble by
the final analysis ensemble members, \(\mathbf{x}^{a(i)}\), are the result of adding each column of \(\mathbf{X}^a\) to \(\bar{\mathbf{x}}^a\)