Chemical state estimation for the middle atmosphere by four-dimensional variational data assimilation: A posteriori validation of error statistics in observation space

Chemical state analyses of the atmosphere based on data assimilation may be degraded by inconsistent covariances of background and observation errors. An efficient method to calculate consistency diagnostics for background and observation errors in observation space is applied to analyses of the four-dimensional variational stratospheric chemistry data assimilation system SACADA (Synoptic Analysis of Chemical Constituents by Advanced Data Assimilation). A background error covariance model for the assimilation of Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) ozone retrievals is set up and optimized. It is shown that a significant improvement of the assimilation system performance is attained through the use of this covariance model compared to a simple covariance formulation, which assumes background errors to be a fixed fraction of the field value. The forecast skill, measured by the distance between the model forecast and MIPAS observations, is shown to improve. Further, an evaluation of analyses with independent data from the Halogen Observation Experiment (HALOE), the Stratospheric Aerosol and Gas Experiment II (SAGE II), and ozone sondes reveals that the standard deviation of ozone analyses with respect to these instruments is reduced throughout the middle stratosphere. Compared to the impact of background error variances on analysis quality, it is found that the precise specification of spatial background error correlations appears to be less critical if observations are spatially and temporally dense. Results indicate that ozone forecast errors of a state of the art stratospheric chemistry assimilation system are of the same order of magnitude as MIPAS observation errors