Image classification is one of the most widely used techniques on remote sensing data. Whereas the actual supervised classification of satellite image is a highly automated process, assembling the training data needed for such classification is anything but automatic. Our research has focused on developing a method to extract the training samples for some classes automatically from the data itself by analyzing the discrete fourier transform of the temporal signatures of every class. The multi- temporal data allows us to characterize objects based on their dynamic processes rather than static properties like color,shape, etc and this is being successfully demonstrated in this research. Other utilities of the time series data like derivation of a season calendar, mapping of cropping practices and finding single and double cropping regions have been demonstrated. The solution to the problem of finding single and multi crop regions enables us to implement policy decisions like Special Economic Zone (SEZ) Act, 2002 which prohibits the use of multi cropping lands for setting up special economic zones.