ASSESSMENT OF CLOUDS CHARACTERISTICS FROM SATELLITE-OBSERVATIONS BY MEANS OF SELF-ORGANIZED NEURAL NETWORKS

The classification of pixels making up a satellite image is seen here, not only in order to cluster or discriminate these pixels, but exactl y as an inversion procedure. The implemented adjustable combination of neural networks (ACNN) technique is built as a combination of self-or ganized neural networks in order to be particularly robust. After a se nsitivity study, this method was considered for the extraction of clou d optical thickness and droplet effective radius from AVHRR imagery. T he validation of this algorithm was conducted in well-documented cases issued from the EUCREX campaigns by comparisons against the CRTVL inv ersion package developed by Nakajima and Nakajima (1995) and also in s itu measurements. (C)Elsevier Science Inc., 1998.