Assessing observation impact in ensemble data assimilation

With the rapid development of exploring techniques, the number and variety of observations have increased significantly. The information of the impact that different observations have on the analyses and forecasts is crucial to better use the observations in a data assimilation system. In this study, an ensemble-based estimation method proposed by Liu and Kalnay (2008) (LK08) using the local ensemble transform Kalman filter is applied to an AGCM global model to assess the observation impact of simulated rawinsondes, cloud drift wind, and satellite retrieved temperature and humidity profiles. With the LK08, it is not necessary to add or remove observations from the assimilation as was done in the traditional data-denial experiments. Our results show that the LK08 procedure can successfully evaluate the impact of each observation on the forecast and the impact values can then be grouped and summed by various subsets of observations that may be of interest. The total observation impact in the Northern Hemisphere is bigger than that in the Southern Hemisphere. The largest impact in the Northern Hemisphere is produced by rawinsondes, whereas it is produced by satellite retrieved profiles in the Southern Hemisphere. Cloud drift wind has largest impact in tropics. The validation analysis shows that the total estimated global observation impact accounts for 70`80% of the value of actual impact and captures the variations of actual impact very well.