Comparison of the Ensemble Adjustment Kalman Filter (EAKF) and Rank Histogram Filter (RHF) with WRF-DART for Two Convective Cases over the Great Plains Region

The mesoscale community Weather Research and Forecasting (WRF) model is used in conjunction with the NCAR Data Assimilation Research Testbed (DART) to perform cycled data assimilation for numerical simulations of convective cases on 23-24 May, 2011 and June 3-4, 2015 over the Southern Great Plains Region. Available conventional and satellite data as well as surface Mesonet observations are assimilated. The Ensemble Adjustment Kalman Filter (EAKF) is compared to the Rank Histogram Filter (RHF) to determine where and why the different filters result in varied analysis and simulations of mesoscale characteristics of the convective system during its initiation and evolution for each case. In particular, the influence of the assumption of non-Gaussian data assimilation was examined with these two filters, since the RHF is designed to handle non-Gaussian prior distributions, while the EAKF does not.

Results show that significant non-Gaussianity is present in all cases. Meanwhile, non-Gaussian regions tend to focus near convection, and to a lesser degree near boundaries and gradients. Compared with RHF analysis, the EAKF analysis contains more non-Gaussianity in its distributions on average.

In general, the EAKF analysis have similar to superior skill compared to the RHF analysis based on RMSEs and biases against observations. However, near non-Gaussian features and at higher resolution, the RHF analysis often resolves these features more accurately than the EAKF analysis does. Further diagnoses have been conducted to examine how the differing assumptions (e.g., non-Gaussian vs. Gaussian) in both filters affects the outcome of the data assimilation. It is noted that the most significant factor that leads to large differences in the analysis is related to deviations from a Gaussian distribution (outliers or bimodal distributions), as was expected. Detailed results will be presented at the conference.