Neural Network based Retrieval of Cirrus Properties

In this work, a method for determining the micro- and macro-physical properties of cirrus clouds from NOAA- AVHRR daylight imagery is presented. The combined use of the radiances measured by satellites and an atmospheric radiative transfer model makes possible to obtain cirrus properties, such as optical thickness, mean effective ice crystal size and temperature, through an inversion method. Due to the complexity of this theoretical model, numerical techniques must be used. In this case, the inversion is performed using an artificial neural network (ANN), which is trained and optimized by genetic algorithms. After the training stage, the ability of generalization of this network and the errors introduced in the procedure are analyzed.