In this work a threshold technique for cloud detection and classification is applied to 9 years NOAA-AVHRR
imagery in order to obtain a cloud climatology of the Canary Islands region (Northeast Atlantic Ocean). Once
the clouds are classified, a retrieval method is used to estimate cloud macro- and micro-physical parameters, such
as, effective particle size, optical thickness and top temperature. This retrieval method is based on the inversion
of the simulated radiances obtained by a numerical radiative transfer model, libRadtran, using artificial neural
networks (ANNs). The ANNs, whose architecture was based on Multilayer Perceptron model, were trained with
simulated theoretical radiances using backpropagation with momentum method, and their architectures were
optimized through genetic algorithms. The global procedure was performed for both day and night overpasses
and, from a set of more than 9000 images, maps of relative frequency were calculated. These results were
compared with ISCCP data for the 21-year period 1984-2004. The relationships between the retrieved cloud
properties and some climate and atmospheric variables were also considered.
In this work a method for determining the micro- and macro-physical properties of oceanic stratocumulus clouds at night-time (when only infrared data are available) is presented. It is based on the inversion of a radiative transfer model that computes the brightness temperatures in NOAA-AVHRR channels 3, 4 and 5. The inversion is performed using an artificial neural network (ANN), which is trained to fit the theoretical computations. A detailed study of the ANN parameters and training algorithms demonstrates the convenience of using the "back propagation with momentum" method. The proposed retrieval, using both uniform and adiabatic models for clouds, was validated using ground data collected in Tenerife (Canary Islands), and a good agreement was obtained in those pixels near the sample site. The convenience of using the adiabatic approximation is discussed.