Date of Award

May 2017

Degree Type


Degree Name

Doctor of Philosophy



First Advisor

Changshan Wu

Committee Members

Alison C. Donnelly, Zengwang Xu, Lingqian Hu, Pinliang Dong


Fusion, Hyperspectral Imagery, Lidar, Pixel-Weighting, Tree Species Identification


Monitoring and management of ash trees has become particularly important in recent years due to the heightened risk of attack from the invasive pest, the emerald ash borer (EAB). However, distinguishing ash from other deciduous trees can be challenging. Both hyperspectral imagery and Light detection and ranging (LiDAR) data are two valuable data sources that are often used for tree species classification. Hyperspectral imagery measures detailed spectral reflectance related to the biochemical properties of vegetation, while LiDAR data measures the three-dimensional structure of tree crowns related to morphological characteristics. Thus, the accuracy of vegetation classification may be improved by combining both techniques. Therefore, the objective of this research is to integrate hyperspectral imagery and LiDAR data for improving ash tree identification. Specifically, the research aims include: 1) using LiDAR data for individual tree crowns segmentation; 2) using hyperspectral imagery for extraction of relative pure crown spectra; 3) fusing hyperspectral and LiDAR data for ash tree identification. It is expected that the classification accuracy of ash trees will be significantly improved with the integration of hyperspectral and LiDAR techniques.

Analysis results suggest that, first, 3D crown structures of individual trees can be reconstructed using a set of generalized geometric models which optimally matched LiDAR-derived raster image, and crown widths can be further estimated using tree height and shape-related parameters as independent variables and ground measurement of crown widths as dependent variables. Second, with constrained linear spectral mixture analysis method, the fractions of all materials within a pixel can be extracted, and relative pure crown-scale spectra can be further calculated using illuminated-leaf fraction as weighting factors for tree species classification. Third, both crown shape index (SI) and coefficient of variation (CV) can be extracted from LiDAR data as invariant variables in tree’s life cycle, and improve ash tree identification by integrating with pixel-weighted crown spectra.

Therefore, three major contributions of this research have been made in the field of tree species classification:1) the automatic estimation of individual tree crown width from LiDAR data by combining a generalized geometric model and a regression model, 2) the computation of relative pure crown-scale spectral reflectance using a pixel-weighting algorithm for tree species classification, 3) the fusion of shape-related structural features and pixel-weighted crown-scale spectral features for improving of ash tree identification.