Large area crop classification in New South Wales, Australia, using Landsat data

Abstract

This paper describes digital crop classification techniques developed for Australian conditions. Using extensive ground truth data, supervized maximum likelihood was used to classify Landsat data from five dates across approximately half a Landsat scene into winter crops and other land cover types. Inaccuracies in the prediction of total crop hectarage and in the classification of an independent test sample of pixels indicated confusion between wheat and barley for all combinations of dates. However, when wheat, barley and oats were considered together against all non-winter crops, the predictions were very accurate and demonstrated the applicability of this method of classification.