Feidas, H. (University of Athens, Department of Physics, Division of Applied Physics)
Cartalis, C. (same affiliation)
In this study a cloud classification is performed systematically on METEOSAT imagery for the case of the flood incident in Greece on 20-22 October 1994. The aim of the classification is the definition of the system characteristics associated with the ocurrence of floods and especially the evolution of the clouds cells associated with heavy rain.
A statistical classification method based on clustering on the three-dimensional histograms is applied to the three channels of the METEOSAT imagery [Visible (VIS), Infrared (IR), Water Vapour (WV)] during the daylight.
The values of the images in visible have been normalised to the maximum value of the day using a sinusoidal formula just to eliminate the effect of the different illumination of the earth during the day. The method of unsupervised classification is performed to the 1200 UTC images for the period of 20 - 22 October 1994 producing after merging one set of spectral signatures. Clustering of the three and two-dimensional histograms, as well as analysis of radiosonde temperature and dew point temperature profiles, supported the correspondence of the final spectral signatures to five categories of cloud type: thick opaque convective, opaque convective, high top (thick and thin cirrus), middle and low top. For demonstrational purposes, images were pseudo-coloured, to allow their convenient use in operational terms. Following supervised classification was applied to the images of the period in hourly frequency, on the basis of the final signature set.
The classified images were found to be supportive of the definition of the clouds cells associated with heavy rain as well as the monitoring of their evolution for this episode.