Cloud mask via cumulative discriminant analysis applied to satellite infrared observations: scientific basis and initial evaluation

Abstract

Abstract. We introduce a classification method (cumulative discriminant analysis) of the discriminant analysis type to discriminate between cloudy and clear-sky satellite observations in the thermal infrared. The tool is intended for the high-spectral-resolution infrared sounder (IRS) planned for the geostationary METEOSAT (Meteorological Satellite) Third Generation platform and uses IASI (Infrared Atmospheric Sounding Interferometer) data as a proxy. The cumulative discriminant analysis does not introduce biases intrinsic with the approximation of the probability density functions and is flexible enough to adapt to different strategies to optimize the cloud mask. The methodology is based on nine statistics computed from IASI spectral radiances, which exploit the high spectral resolution of the instrument and which effectively summarize information contained within the IASI spectrum. A principal component analysis prior step is also introduced, which makes the problem more consistent with the statistical assumptions of the methodology. An initial assessment of the scheme is performed based on global and regional IASI real data sets and cloud masks obtained from AVHRR (Advanced Very High Resolution Radiometer) and SEVIRI (Spinning Enhanced Visible and Infrared Imager) imagers. The agreement with these independent cloud masks is generally well above 80 %, except at high latitudes in the winter seasons.