Classification of IASI inhomogeneous scenes using co-located AVHRR data

Abstract

The Infrared Atmospheric Sounding Interferometer (IASI) , due to fly on the EPS Metop satellites from 2006 onwards, will produce hyperspectral radiance spectra, profiles of temperature and humidity at high vertical resolution in addition to columnar trace gas amounts, skin temperature, surface emissivity, and cloud products. Since IASI is not an ideal interferometer, any inhomogeneity within the instrument's field of view (FOV) changes the self-apodisation and thus modifies the spectral response. Clouds and surface variability are likely sources of such inhomogeneities that need to be known in order to correct for the changed spectral response and perform the atmospheric retrievals to the specified accuracy. The operationally generated AVHRR (Advanced Very High Resolution Radiometer) scenes analysis is used to identify the surface or cloud types present in each IASI FOV. However, although statistically averaging the scenes analysis results to obtain mean estimates of cloud coverage and scene type within the IASI pixels increases the probability of correct classification, the geometric distribution of each scene is lost. A radiance analysis is performed on the co-located AVHRR pixels to identify localised areas of similar radiance clusters within the IASI pixels, the assumption being these clusters correspond to different scene types. Similar clusters of known scene type are then generated from the AVHRR scenes analysis data and matched to the clusters generated from the independent radiance analysis in order to classify scenes in IASI fields of view.