High-resolution infrared sounders, such as the Infrared Atmospheric Sounding Interferometer (IASI) on the current MetOp series of satellites, produce several orders of magnitude more data per location than previous instruments used in operational retrieval and data assimilation schemes. Using the full spectrum (8641 channels for IASI) is impractical and a common approach is to identify a subset of channels which, ideally, conveys the most information on the target parameters (e.g. atmospheric temperature and water vapour) but using a relatively small number of measurements.
Representing the problem as a one-dimensional retrieval, optimal estimation provides an efficient framework for channel selection, and is the basis of several current schemes. However, while modelling the propagation of random (spectrally uncorrelated) errors into the retrieval, the standard algorithm does not allow for spectrally correlated errors, particularly arising from the radiative transfer modelling, which are often the limiting factor in retrieval accuracy. Such errors are either ignored or represented only approximately during the selection.
This article describes a modification to the standard algorithm which allows spectrally correlated errors to be properly modelled, and quantified, within the channel selection process. Comparing the results with an established selection scheme shows that significant improvements can be obtained when retrieving temperature regarding water vapour as an error term, but are less dramatic when both are retrieved together. The concept of ‘total’ information available from an IASI spectrum is also re-assessed.