Influence of background error covariances on assimilating hurricane inner-core observations in HWRF nested domains with the GSI hybrid data assimilation system

A series of data assimilation experiments have been conducted to examine the influence of background error covariances on assimilating hurricane inner-core observations in a hurricane weather research and forecasting (HWRF) model. The NCEP gridpoint statistical interpolation (GSI) data assimilation system is used. Various configurations of experiments are performed with assimilating inner-core Tail Doppler Radar (TDR) observations. Specifically, the vortex spin-down problem in current HWRF and GSI data assimilation system are identified and characterized.

Hurricane Earl (2010) at its mature phase is used as a case study. The first group of experiments include: 1) No inner-core TDR data assimilation (while other routine observations are assimilated), 2) GSI 3DVAR-EnKF hybrid (GSI hybrid) data assimilation with inner-core TDR data, 3) GSI 3DVAR data assimilation with TDR data. Both no-cycle and cycled data assimilation are performed. Results indicate a strong vortex spin-down problem during the first 2-h of forecasts (after the vortex initialization and data assimilation) in all experiments. With the GSI hybrid assimilation of TDR inner-core data, the HWRF forecast slightly outperforms other experiments in terms of the intensity forecast. However, the GSI hybrid system still cannot overcome the spin-down problem.

The analysis increments, vortex structures before, after data assimilation experiments and during the vortex spin-down periods, have been examined in details. Notable vortex structure (in terms of vortex size, structural feature and distribution of the wind, warm core etc.) changes were found during the vortex spin down period. In particular, the diagnoses indicate that the background covariance term, generated from the global ensemble forecasts in GSI hybrid system, results in large-scale features in the analysis increments near and inside of the hurricane vortices. These large-scale features have been quickly adjusted during the vortex spin-down period.

Additional numerical experiments are performed to use the ensemble forecasts at HWRF native domains, instead of the global ensemble forecasts, in the background term of the GSI 3DVAR-EnKF hybrid data assimilation system. It is found that the use of regional ensemble forecasts provides reasonable background error terms, thus improving the HWRF analysis and forecasts and also mitigating the vortex spin-down problem.