A hybrid ensemble Kalman filter approach to data assimilation in WRF / DART

A hybrid ensemble Kalman filter, previously tested in the Lorenz three-variable system and a two-dimensional shallow water model, is now explored in WRF / DART. This hybrid ensemble Kalman filter approach effectively combines the ensemble Kalman filter (EnKF) and observation nudging to achieve a more gradual and flow-dependent data assimilation. It computes the hybrid nudging coefficients from the EnKF error covariances. It extends the specified influence function used in nudging to one based on flow-dependent error correlations and observation errors. This hybrid EnKF can also transform the gain matrix of the EnKF into additional off-diagonal statistical terms to complement the standard diagonal relaxation terms in the model's predictive equations to assist in the data assimilation process.

The hybrid EnKF approach is tested using a Cross Appalachian Tracer Experiment case from September 1983 (CAPTEX-83). It features a mid-latitude cyclone with frontal systems and precipitation, and observed tracer data. The CAPTEX-83 case is used to evaluate the potential for using this hybrid EnKF approach to perform a more gradual, flow-dependent assimilation of standard WMO data compared to the use of continuous nudging or intermittent EnKF in a real-data case. The CAPTEX-83 case is widely used for air-quality and regional transport studies. In addition to computing standard closeness-of-fit statistics against the observations to evaluate and compare the data assimilation methods, an independent verification of the data assimilation approaches is also performed using the observed tracer data. The WRF experiment results are used in the Second-Order Closure Integrated Puff (SCIPUFF) atmospheric transport and dispersion model to predict surface tracer concentrations that are then verified against the observed surface concentration data.

Preliminary results show the hybrid EnKF producing the best results for the tracer data, followed by observation nudging and then the EnKF. This demonstrates that use of continuous nudging and intermittent EnKF together produces analyses that outperform either method applied separately.