In supervised classification, the analyst guides the classification procedure by providing representative samples (also known as reference or training data) for each of the predefined classes (Lillesand et al., 2004). In the case of crop mapping, this means to indicate representative fields for each crop type that one wants to be identified. It is good practice to divide one’s reference data into two parts; one set for training the classifier and another set for validation purposes. This approach will allow to make a value statement on the quality of the developed classifier.
Supervised classification techniques are recommended when a field campaign has been carried out to map representative crop types. It gives the analyst more control over the process and more confidence in the results. Supervised techniques give more accurate classification results than unsupervised methods (Tso and Mather, 2009). In cases where no prior knowledge of the location of representative classes exists, an unsupervised technique has to be adopted.
Since the STARS project undertook substantial field campaign work to generate reference data, supervised classification techniques were employed.