ASSESSING THE IMPACT OF INCORRECT OBSERVATIONAL COVARIANCE MATRIX OVER RETRIEVAL: METHODS AND APPLICATION TO IASI DATA

The paper addresses the effect of an incorrect specified observational covariance matrix on the retrieval of atmospheric parameters from high spectral resolution infrared radiances. The case of a non-diagonal covariance matrix approximated with its diagonal is dealt with. The problem is of interest to IASI (Infrared Atmospheric Sounding Interferometer) which, because of apodization, is characterized by a non-diagonal covariance matrix. However, the diagonal alone rather than the full matrix is normally used in data assimilation and retrieval schemes because of numerical and computational efficiency. The problem will be analysed in its formal, mathematical aspects through the help of the linear approximation. In addition, a series of retrieval exercises will help to draw more general conclusions. We have found that the incorrect use of the diagonal instead of the full covariance matrix can affect the spatial vertical resolution of the retrieval and can lead to instability in the final solution, especially in the lower troposphere.