Application of Hybrid Variational-ensemble Assimilation Method in FY-3A Satellite Microwave Data

A variational-ensemble data assimilation system for FY-3A satellite data was constructed using the WRF-Hybrid method. In order to evaluate the capability of assimilation system, single observation experiments and observation system simulation experiments and real data assimilation experiments for FY-3A MWHS and MWTS sensor data were designed for the Beijing “7.21” extreme precipitation event. The results show that the system can assimilate satellite data correctly, and its background error covariance has obvious flow-dependent characteristics. The results of the OSSEs show that the RMSEs of the analysis field for assimilation experiments with the "real" field are less than those of the background field, which indicates that the system can assimilate satellite data effectively.

Sensitivity of the system to key parameters including the numbers and types of ensemble samples, hybrid coefficients and the types of satellite microwave data was tested in sensitivity experiments. For the WRF-Hybrid method, the number of ensemble samples affects the size and distribution of ensemble covariance, which affects the effect of assimilation. For this reason, sensitivity experiments were designed to reveal the effect of the number of ensemble samples on assimilation. In the experiments, values of 50, 68, 100, 150 and 168 were used for the number of ensemble samples. Values of 50,100 and 150 for ensemble samples were obtained by using the breeding of growing modes (BGM). Values of 68 for ensemble samples include 18 physical ensemble samples and 50 BGM initial ensemble samples. Values of 168 for ensemble samples include 18 physical ensemble samples and 150 BGM initial ensemble samples. And the results show that increasing the number of initial ensemble members has little influence on the assimilation result, but increasing physical ensemble members clearly increases the assimilation ability.

Ensemble covariance in WRF-Hybrid method depends on ensemble samples. Ensemble samples in the experiment were obtained by using BGM and ensemble transform Kalman filter (ETKF) and super-ensemble (SUP) including the initial value ensemble samples and multi-physics process model forecast samples. The distribution and size of ensemble perturbation have been quantitative evaluated. By a single point observation experiment and a real data assimilation experiment, the characteristics of the ensemble covariance which are produced by different ensemble samples and its impacts on model forecast accuracy have been revealed. The main conclusions are as follows: Among BGM, ETKF and SUP ensemble samples, ensemble covariance constructed by SUP ensemble samples show more significant flow-dependent characteristic, and assimilation increments perform more effective improvement in analysis field.The effect on the assimilation is affected synthetically by background field and background error covariance. A high level of ensemble mean field and the responsibility of ensemble spread can produce background field and background error covariance which give the closest agreement with the real case.

The flow-dependent characteristic of hybrid covariance depends on hybrid coefficients of ensemble covariance. Different variables and circulation system forecasts are selective about the optimal hybrid coefficient. When hybrid coefficient is small, hybrid covariance show homogeneous and isotropic characteristics，and generate a large-scale assimilation increments which effectively improve large-scale circulation system in background field; When hybrid coefficient is large, hybrid covariance show more obvious flow-dependent characteristic. The sensitivity experiments demonstrate that assimilation system shows strong assimilation ability when the hybrid coefficient is 0.5. Because background error covariance has the multi-scale characteristic, the proportion of static and ensemble covariance should be very closely coordinated with the proportion of large-scale background error and the small-scale background error. Only in this way can the optimal increment be obtained.

The FY-3A MWHS and MWTS and NOAA-16 AMSU-A and AMSU-B microwave data were used for evaluating the capability to the assimilation system. After quality control and bias correction, the systematic errors of FY-3A microwave data and NOAA-16 microwave data were effectively reduced. The distribution tends to be reasonable, and the deviation of the two satellite data is equivalent. Compared to assimilation of NOAA-16 AMSU-A and AMSU-B data, the forecast results of assimilation of FY-3A MWHS and MWTS data are better for heavy rain and extremely heavy rain.