Abstract
Knowledge of the extent and degree of wooded vegetation cover in the arid parts of Australia is essential for land-holders and management agencies. The balance between wooded and ground-cover vegetation is important to livestock production and landscape health. Adequate mapping of changes in wooded vegetation cover allows the assessment of its expansion and contraction as input for improved management of production and conservation. The aim of this study was to develop a method to accurately map the extent and degree of wooded tree and shrub cover across an area of arid rangeland in central Australia. Its open and sparse distribution throughout the landscape and its adaptation to an arid environment present challenges to obtaining both representative field measurements and scale appropriate remotely sensed imagery. Recent advancements and access to high spatial resolution satellite imagery provide opportunities for improved mapping. The rapid development of Remotely Piloted Aircraft Systems (RPAS) or drones also provides further opportunities to improve the accuracy of field measurements used in the classification of wooded vegetation. An optimised machine-learning classification was developed using high resolution Planet Dove cube-sat and Sentinel2 imagery and compared to medium resolution Landsat8 imagery. An efficient method of collecting plot scale (ha) wooded vegetation cover estimates for the training and assessment of the satellite image classification was also developed using the RPAS. It was comparable to other field based measurements. The results of the classifications showed a moderate degree of accuracy in distinguishing wooded cover from non-wooded cover, highest with the Planet Dove imagery. An improved accuracy in distinguishing between wooded cover classes was also seen in the Sentinel2 classification. The mapping and subsequent monitoring of wooded vegetation in these landscapes has been shown to be improved with higher resolution satellite imagery.
NSTL
Elsevier