Abstract: Accurate specification of the background error covariance matrix (B) is a fundamental prerequisite for a successful data assimilation scheme. However, most of operational variational data assimilation systems still rely on a static, climatological representation of the B, and thus implicitly renounce the representation of the flow-dependent covariance errors. To overcome this limitation, an Ensemble Data Assimilations (EDA) approach and spherical wavelet are used in this study to specify the flow-dependent background error variance and covariance respectively. In this model, both the background error variance and the vertical localization covariance matrix are replaced by that estimated from ensemble samples. Meanwhile, the EDA system proposed here is used to provide estimates of day-to-day background error statistics, which can represent the current meteorological situation through an ensemble of 10 lower-resolution four dimensional variational data assimilation(4DVar) analysis cycles that make use of perturbed observations, perturbed sea surface temperature (SST) fields and perturbed model physical tendencies. However, a key problem is that the sample variances computed from the EDA are affected by the random, which needs to be addressed. In this paper, objective filtering has been implemented due to its effectiveness in extracting the statistically significant part of the signal. This study also evaluates the ability of the EDA in representing flow-dependent background-error variances. In particular, the ensemble-based variances are examined in the case of hurricane KEBI, which is the 9th most intense hurricane that occurred in August 2013. Results show that, the EDA system with 10 members is able to correct errors in large scale structures associated with tropical cyclogenesis and produce a realistic estimation of the background error that varies with weather condition effectively and accurately. In this sense, the objective filtering technique provides a useful indication of the spatial scales the ensemble is able to resolve in a statistically robust way. Another interesting aspect is that the use of flow-dependent background error covariance shows some positive impacts on the analysis and forecast of this tremendously varying weather processes.