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 exactly as an inversion procedure. The implemented adjustable combination of neural networks (ACNN) technique is built as a combination of self-organized neural networks in order to be particularly robust. After a sensitivity study, this method was considered for the extraction of cloud optical thickness and droplet effective radius from AVHRR imagery. The validation of this algorithm was conducted in well-documented cases issued from the EUCREX campaigns by comparisons against the CRTVL inversion package developed by Nakajima and Nakajima (1995) and also in situ measurements.