A neural network approach to cloud classification from multi-temporal satellite imagery

In response both to the general lack of automatic methods to analyse the increasing amount of satellite data, and to the availability of multi-temporal information at high temporal resolution (e.g. Meteosat, 30 minutes), a new artificial neural network (ANN) method for classifying clouds has been developed. A recently developed cloud tracking method, utilising a Hopfield neural network, is used to acquire new dynamic cloud parameters from satellite image sequences. These parameters are analysed, and their contribution towards accurate classification is discussed.