Impacts of Model Errors and Ensemble Initiation on Mesoscale Ensemble-Based Data Assimilation

The ensemble-based data assimilation techniques, which estimate background error covariance through an ensemble of short-term forecasts, have been recently applied to various dynamical systems with various scales of interest. A recent study by the authors has demonstrated the feasibility of using an ensemble Kalman filter (EnKF), for mesoscale and regional-scale data assimilation through various observing system simulation experiments (OSSEs) assuming a perfect forecast model with perfect ensemble initiation. This current study seeks to explore the performance of the EnKF in more realistic scenarios when real data are assimilated, which include the presence of significant model errors and the ensemble initiation with imperfect statistics. The EnKF is implemented in the mesoscale model MM5 framework to assimilate synthetic sounding and surface observations with typical temporal and spatial resolutions derived from the truth simulation. The EnKF is applied to two cases with different flow regimes. One is the ‘surprise' winter snowstorm of 24-26 January 2000 and the other is a long-lived warm-season mesoscale convective vortex (MCV) event of 10-13 June 2003.

The model errors considered here include generating the reference (truth) simulations from using different physical parameterization schemes and using different model grid resolutions. Three different ensemble generation techniques are used to generate the initial ensemble forecasts, which include (1) pure random (white) noise, (2) balanced random perturbations, and (3) the combination of both. In most of the imperfect scenarios tested, the EnKF is found to be quite resilient with 30 to 60% improvement over the pure ensemble forecast errors, though the use of an imperfect model or imperfect ensemble initialization can significantly degrade the filter performance. Similar to the perfect scenario, most of the error reduction comes from larger scales. The scale- and dynamic-dependence of the EnKF leads to its varying performance for the two cases examined. It is also found that, in the presence of model error, using a combination of different parameterization schemes in the ensemble forecast can significantly improve the filter performance.