Evaluation of the Potential for Detection and Classification of Ailanthus altissima (Tree of Heaven) Using LiDAR Data

This thesis compares methods for delineating and classifying the invasive, exotic tree Ailanthus altissima (tree of heaven), using attributes derived entirely from light detection and ranging (LiDAR) data. The accuracy of two image segmentation methods: 1) Tree variable window program (TreeVaW) and 2) watershed segmentation, and three classifications schemes: 1) classification and regression trees (CART) 2) artificial neural networks (NN) and 3) support vector machines (SVM) are compared. I found that generally the watershed segmentation method produced better segmentation results than the TreeVaW segmentation method, and that the CART classification was the most accurate overall classifier, although the SVM classification produced the most accurate Ailanthus species classification. The factors that are most important in influencing the segmentation and classification accuracies are the point density of the LiDAR data, the level of tree-crown penetration by the LiDAR laser pulses, and the quality of the canopy height model derived from the LiDAR data point cloud. CART and SVM classification, together with watershed segmentation are optimal methods of identifying Ailanthus altissima trees from LiDAR data.