Frontiers in Using LiDAR to Analyze Urban Landscape Heterogeneity

Light Detection and Ranging (LiDAR) technology has facilitated extraordinary advances in our ability to remotely sense precise details of both built and natural environments. The inherent complexity of urban landscapes and the massive data volumes produced by LiDAR require unique methodological considerations for big data remote sensing over large metropolitan regions. The heterogeneous landscapes of the rapidly urbanizing Charlotte Metropolitan Region of North Carolina provided an ideal testing ground for developing methods of analysis for urban ecosystems over large regional extents, including: (1) fusion of LiDAR digital surface models (DSMs) with Landsat TM imagery to balance spatial resolution, data volume, and mapping accuracy of urban land covers, (2) comparison of LiDAR-derived metrics to fine grain optical imagery -- and their integration -- for detecting forest understory plant invaders, and (3) data reduction techniques for computationally efficient estimation of aboveground woody biomass in urban forests.;In Chapter 1, I examined tradeoffs between potential gains in mapping accuracy and computational costs by integrating DSMs (structural and intensity) extracted from LiDAR with TM imagery and evaluating the degree to which TM, LiDAR, and LiDAR-TM fusion data discriminated land covers. I used Maximum Likelihood and Classification Tree algorithms to classify TM data, LiDAR data, and LiDAR-TM fusions. I assessed the relative contributions of LiDAR DSMs to map classification accuracy and identified an optimal spatial resolution of LiDAR DSMs for large area assessments of urban land cover. In Chapter 2, I analyzed combinations of datasets developed from categorized LiDAR-derived variables (Overstory, Understory, Topography, and Overall Vegetation Characteristics) and IKONOS imagery ( Optical) to detect and map the understory plant invader, Ligustrum sinense, using Random Forest (RF) and logistic regression (LR) algorithms, and I assessed the relative contributions of sensors and forest landscape structures. I compared the top performing models developed using RF and LR and used the best overall model to map the distribution of L. sinense occurrence across the urbanizing forest landscapes of the region. In chapter 3, I examined the effects of LiDAR point density and landscape context on the estimation of biomass (of general Urban Forest and of three specific Forest Types) using multiple linear regression. I compared biomass estimation accuracies of the Urban Forest and Forest Type models and between the top-performing models of these two Forest categories. For the effect of landscape context, I quantified the degree to which the presence of built development influenced biomass estimation, and I analyzed the effect of canopy stratification on the estimation of biomass.;A unifying theme of my dissertation is to advance LiDAR analytics for accurate and detailed estimation of urban landscape heterogeneity over large regional extents. The results of the three studies suggest that establishing optimal resolution and point density for LiDAR data is a highly effective method of pursuing large area studies of urban landscape heterogeneity, and the fusion of LiDAR-derived variables and multispectral data is beneficial in some applications such as improving class discrimination of spectrally similar land cover types. Finally, the direct measurement of forest understory and overstory structure through LiDAR has proven valuable for the study of complex and heterogeneous ecosystems like urban forests.