Classification of Vegetation Communities in the Tropical Savannas of Northern Australia Using Airsar Data

Radar remote sensing offers great potential for resource management in tropical regions. This paper outlines a processing methodology for AirSAR data for land cover classification using standard image processing software. The side looking nature of the instrument introduces significant variation in incidence angle from an airborne platform. The changes introduced by variation in incidence angle are dependent on the land cover and can, therefore, not be removed by mathematical modeling unless the exact ground cover composition is known. The proposed methods for the removal of this effect are based on the statistical comparison of lines of data with constant incidence angle. It is shown, for the vegetation communities in a coastal tropical savanna landscape in Australia’s Northern Territory, that a successful correction method for the effects of incidence angle variation can be implemented. The elimination of backscatter dependence on incidence angle allows the qualitative discrimination of land cover types using a supervised technique. A maximum likelihood classification of the data achieved an overall accuracy of 87.2% for six land cover types. The separation of classes is based on structural rather than species differences. Subdivision of the Eucalypt forest class derived from the supervised approach delineates distinct classes but further research is required to determine the biophysical parameters of vegetation determining the resulting stratification.