Study Of Assimilating Doppler Radar Data Using Hybrid EnSRF-En3DVar Method

Variational data assimilation and ensemble Kalman filter (EnKF) are two major methods of current data assimilation (DA) techniques. As the relative advantages and disadvantages of variational and EnKF being extensively discussed, a new method called hybrid ensemble-variational DA method which blends the advanced features of both ensemble and variational methods and overcomes their shortcomings is proposed and gradually becomes an important research direction of DA. Further studies demonstrat the potential advantages of the hybrid method. In the hybrid DA system, ensemble forecast members can realistically infer the flow-dependent background covariance error statistics and the updated forecast perturbations provide initial conditions for the next ensemble forecast cycle, therefore choosing a suitable ensemble purterbations’ updated scheme becomes the key to the success of hybrid DA system.As a pilot study applying EnSRF to update ensemble perturbations, A hybrid ensemble square root filter and three-dimensional ensemble-variational (EnSRF-En3DVar) DA system is developed on WRF and explored its potential for assimilating radar observations. Three sets of experiments are conducted to test this system on typhoon "saomai". The first and the second are choosing different weight coefficient of ensemble covariance and covariance horizontal localization scale, the third set of experiments is assimilating radar radial velocity(Vr) and/or reflectivity(Z). Finally, while previous studies which focus on radar DA with hybrid method always use "perturbed observation method" to update forecast perturbations, the methods to generate the ensemble perturbations are also be compared.The results show that:(1) Hybrid single observation tests show the impact on covariance horizontal localization scale. The value 20,60,200,600 km for horizontal localization are tested and found that the value of 600 km and 200 km allow a single radial velocity observation to influence the majority of the domain, while 20 km constrains the analysis increments to a more local vicinity around the observation. The value 60 km shows the most reasonable increment.(2) Observing system simulation experiments (OSSE) in hybrid EnSRF-En3DVar DA system show that experiments with 30 ensemble members and a relative weighting (0.25,0.5,0.75) for the ensemble covariance lead to better analyses than the corresponding 3DVar. When using pure ensemble covariance, En3DVar hybrid system performs the worst while a relative weighting (0.75) for the ensemble covariance gets the best analysis.(3) The storm structures can be established reasonably well when assimilating radial velocity and/or radial reflectivity in hybrid EnSRF-En3DVar DA system. The further analysis shows that assimilating Vr and Z help improve the initial analysis of humidity and wind fields, but assimilating Vr only help improve the analysis of wind field most.(4) The method to generate the ensemble perturbations also be compared, the "perturbed observation" method gives smaller ensemble spread and needs more expensive computational cost than the EnSRF method, so the EnSRF method is superior to "perturbed observation" method to generate the ensemble perturbations.