Pixel-based image classification to map vegetation communities using SPOT5 and landsat TM in a northern territory tropical savanna, Australia

Traditional techniques to map vegetation communities are by aerial photography interpretation and intensive field sampling. Semi-automated methods, including pixel and object-based image classification, demonstrate potential to accurately map vegetation communities, however, there is a lack of comparative research. This study is a component of a broader research project that compares several techniques and image datasets to map vegetation communities. We evaluated a pixel-based supervised image classification using the Maximum Likelihood Classifier and floristic and structural field data applied to SPOT5 multispectral and Landsat5 TM. The study area covered a subset of Bullo River Station in the Top End of the Northern Territory, Australia. Twenty two vegetation communities were classified based on 411 full floristic and structural plots. Class separability averaged 1.94 and 1.42 for Landsat5 TM and SPOT5 respectively. Overall accuracy ranged from 30-53% for 1:25000 and 1:100000 spatial scale products