Accounting for Infrared Sounder Correlated Observation Errors in the GSI

Satellite observations in the infrared band are operationally assimilated at NCEP and are a vital part of the data assimilation system. The Gridpoint Statistical Interpolation (GSI) uses prescribed observation errors for infrared sounders and assumes that the observation errors of different channels are uncorrelated. This presentation uses the Desroziers diagnostic to estimate the observation errors and their inter-channel correlations for Infrared Atmospheric Sounding Interferometer (IASI) and Atmospheric Infrared Sounder (AIRS) observations used in the GSI assimilation system. The new error covariances are then used in a subsequent analysis scheme, and the diagnostic recomputed, until convergence. Accounting for these inter-channel correlations inevitably increases the cost function and can potentially degrade the minimization.  Thus an additional treatment of the covariance matrix, reconditioning, is necessary.  This presentation details the estimation and conditioning of the correlated error covariance matrices and compares different techniques for reconditioning.  The full covariances of IASI and AIRS have been tested in a GSI assimilation experiment.  The results show a significant impact when accounting for the cross-correlations; the new covariance improves the fit to humidity and temperature observations, as well as the fit to passive IR humidity channels.