Land cover classification in Portugal with multitemporal AWiFS images: a comparative study

Three AWiFS scenes from three intra-annual time periods were explored for land cover mapping of mainland Portugal in an annual operational basis. Two different classification approaches were tested: a more conventional parametric classifier, the maximum likelihood classifier, and a nonparametric classifier, a decision tree. Several tests were designed to evaluate the most suitable classification approach, training sample size effect on the classification accuracy and the images's ability for land cover mapping of mainland Portugal. Overall accuracy values achieved were low regardless the classification approach and training sample size, which suggests that AWiFS images are not suitable for land cover mapping the Portuguese landscape. Results also reveal that decision tree classifier is advantageous relatively to maximum likelihood classifier concerning training size and its characteristics.