Assessing the potential of sub-pixel classification in a mixed conifer-broadleaf forest

New Zealand´s indigenous mixed conifer-broadleaf forests are undergoing significant changes in composition, often due to the effect of introduced herbivores. The change can be broadly characterised by a shift in the fraction of four forest groups: emergent conifers, high canopy broadleaf trees, low canopy broadleaf trees, and tree ferns. The authors present an initial evaluation of a sub-pixel classification algorithm for classifying these forest groups from Landsat TM imagery. The authors conclude that the algorithm has potential for mapping the fraction of the groups in mixed conifer-broadleaf forests. This is supported by strong visual similarities between the mapped forest classes and the spatial distribution and frequency of the classified forest groups, and by the good quantitative agreement between the relative amounts of each indicator group in each forest class. The study also highlights the difficulties in obtaining training data and verifying the results of sub-pixel classification in a heterogeneous environment