Adoption of airborne LiDAR data and high spatial resolution satellite imagery for characterisation and classification of forest communities : tests applications in Australian cool temperate rainforest environment.

High resolution spatial data, including airborne LiDAR data and newly available WorldView-2 satellite imagery, offer excellent opportunities to develop new and efficient ways of solving conventional problems in forestry. Those responsible for monitoring forest changes over time relevant to timber harvesting and native forest conservation see the potential for improved documentation from using such data. However, the transfer of new remote sensing technologies from the research domain into operational forestry applications poses challenges. One of the key challenges is the development of a comprehensive procedure which involves deployment of these new remote sensing data to create forest mapping products that are comparable (or superior) in accuracy to conventional photo-interpreted maps. The last decade has witnessed an increase in interest in the application of airborne LiDAR data and high spatial resolution satellite imagery for tree species identification and classification. The research investigations have focused on open forests, and conifer or deciduous forests which are even-aged and of relatively homogenous structures. The suitability of these new remotely sensed data for delineating the structure of complex forest types, particularly for Australian cool temperate rainforest and neighbouring uneven-aged mixed forests in a severely disturbed landscape has hitherto remained untested. This thesis presents ways of processing airborne LiDAR data and high spatial resolution WorldView-2 satellite imagery for characterisation and classification of forest communities in the Strzelecki Ranges, Victoria, Australia. This is a highly disturbed landscape that consists of forestry plantations and large stands of natural forest, including cool temperate rainforest remnants. The k-means clustering algorithm was applied to nonnalised LiDAR points to stratify the vertical forest structure into three layers. Variables characterising the height distribution and density of forest components were derived from LiDAR data within each of these layers. These layer-specific variables were found to be effective in forest classification. Individual trees, including locations and crown sizes, were identified from a LiDAR-derived canopy height model using the TreeVaW algorithm. Augmentation of infonnation extraction from LiDAR data for tree species identification by inclusion of LiDAR intensity data was then tested using statistical analysis techniques. This study demonstrated the contribution of LiDAR-derived intensity variables to the identification of Myrtle Beech (Nothofagus cunninghamii -the dominant species of the Australian cool temperate rainforest in the study area) and adjacent tree species -notably, Silver Wattle (Acacia dealbata) at the individual tree level. Nonparametric classifiers including support vector machines (SVMs) and decision trees were employed to take full advantage of the rich set of infonnation derived from the LiDAR and WorldView-2 imagery data for further improvement in classification accuracy. It is evident that the SVMs have significant advantages over the traditional classification methods in tenns of classification accuracy. Cool temperate rainforest and adjacent forest species were successfully classified from airborne LiDAR data and WorldView-2 satellite imagery using a decision tree approach to object-based analyses in eCognition software. The improvements in results from the methods developed in this study strongly warrant the operational adoption of airborne LiDAR data and high spatial resolution satellite imagery in the management of Australia's forestry resources.