An examination of ensemble filter based adaptive observation methodologies

The type of adaptive observation (AO) schemes of interest in this paper are those which make use of an ensemble forecast generated at a given initial time. The ensemble forecast can be used to quantify the influence of hypothetical observational networks on forecast error covariances. The ensemble transform kalman filter (ETKF) scheme is an example of such a scheme and is used operationally at the National Centers for Environmental Prediction (NCEP). A Bayesian framework for ETKF schemes is developed in this paper. New ETKF AO schemes that make use of covariance localization (CL) are introduced. CL is a technique used to alleviate problems due to sampling errors when estimating covariances from finite samples. No previous study has developed ETKF schemes that make use of CL. A series of observing system simulation experiments (OSSEs) in the non-linear Lorenz 1996 model are used to develop a fundamental understanding of ETKF methods. The OSSEs simulate the problem of choosing observations in a large data void region, to improve forecasts in a verification region located within the data void region. The results demonstrate the important role that techniques for alleviating problems due to sampling errors play in improving the performance of ensemble-based AO techniques