CANOPY PROFILING FOR VEGETATION MAPPING IN SOUTH-WESTERN AUSTRALIAN FORESTED ECOSYSTEMS
Keywords: LIDAR, Vegetation, Geomorphology, Object, Classification, Recognition
Abstract. Anthropogenic climate change is already impacting native vegetation world-wide. Thus accurate mapping of current vegetation condition is necessary for benchmarking and conservation planning. We examine the potential for the mapping of native vegetation of forested ecosystems in south-western Australia using LiDAR data. Airborne LiDAR (distance between data points 1.2 m) and RGB imagery was acquired with a discrete 4-return Leica ALS 50-II system in April 2011 and vertical canopy profiles determined in Boyagin Nature Reserve. Elevation, slope and geomorphological descriptions of the terrain in combination with vertical canopy profiles based on presence/absence of returns within voxels were derived from the LiDAR data and processed at a spatial resolution of 4.0 meters. Based on these profiles, crown height and depth, ground cover, mean intensity of crown returns, presence of understory and number of vegetation layers were determined for each grid cell. Unsupervised classification revealed distinctive canopy profiles. Vegetation is strongly linked to geomorphology in this old landscape. Thus Kwongan shrubland occurs on the plateaus, Allocasuarina heugeliana woodland on the fringes of rock outcrops, Eucalyptus astringens and E. accedens woodland on breakaways and E. wandoo and Corymbia calophylla woodland in more fertile valley systems. The vegetation types present within distinctive spatial clusters were determined in two field visits. Vegetation types were mapped with an object-based image analysis approach at geomorphological, vegetation and tree scales using the geomorphology of the terrain and structural, textural and reflective characteristics of the canopy. All vertical profiles identified were present on each geomorphological unit. Tree species with a distinctive vertical profile such as Eucalyptus astringens and Allocasuarina heugeliana were defined and distinguished in combination with object-based geomorphological and spatial characteristics. Vegetation types were mapped accurately with a kappa coefficient of 0.80. We conclude that vertical profiles and geomorphology of the terrain derived from commercially available discrete return LiDAR systems provides a valuable means to benchmark the mapping of native vegetation in the forested ecosystems of south-western Australia.