Study On The Dual-Resolution Approach For Ensemble Kalman Filter

Recently, Ensemble Kalman Filter (EnKF) has become a research focus in the current data assimilation field. Comparing with other assimilation methods, this approach has the advantage of being able to obtain the flow-dependent background error covariance by using short-term ensemble forecasts. However, the huge computational cost of producing a forecast ensemble becomes the primary challenge for its application into high-resolution mesocale models under an operational environment. In order to reduce the computational cost, an updated dual-resolution strategy for ensemble square-root filter (EnSRF) algorithm is developed based on the dual-resolution EnSRF algorithm proposed by Gao et al,(2007). And serial experiments using the mesocale WRF model are performed. The results show that this approach not only has low computational cost but also can improve the performance of the dual-resolution strategy.Firstly, we examine the trait and flaw of the dual-resolution EnSRF algorithm. As it is shown, the main virtue of this approach is that it can save the computation cost. Moreover, the performance of its ability is good. However, there is a difference in a sense between the forecast error covariance from the high-resolution model and the one interpolated from low-resolution model. It is possible and necessary to adjust the latter error covariance.For this purpose, a research on the structures of forecasted error covariance from high-resolution and low-resolution model is carried out to find the relation between them , using the simulated data. Then, according to this relation, a function is obtained to adjust the latter error covariance and then we compare the results with the covariance from the high-resolution model. It shows that the Boltzmann function performs excellently in fitting the variance between distance and the ratio of average forecast error covariance from the two. But it should vary according to different model levels and variables. Besides, the adjusted standard deviation and covariance are closer to the estimated ones from the high resolution model. At last, a new approach called adjusted dual-resolution is proposed using the adjusted covariance. Sensitivity experiments are performed to examine the aspects related to observation variables and data density .It appears that the capability of adjusted dual-resolution method is better than the original one, especially when more observation variables are involved. Moreover, this predominance is also significant in consecutive assimilation cycles. In addition, it is feasible that the adjusted relation is of applicability in the later analysis. The result of comparison with high-resolution EnSRF shows that adjusted dual-resolution algorithm is slightly poor than standard full EnKF method, but the former has a much lower computational cost under the same running environment.