Advances in measuring forest structure by terrestrial laser scanning with the Dual Wavelength ECHIDNA® LIDAR (DWEL)

Abstract

Leaves in forests assimilate carbon from the atmosphere and woody components store the net production of that assimilation. Separate structure measurements of leaves and woody components advance the monitoring and modeling of forest ecosystem functions. This dissertation provides a method to determine, for the first time, the 3-D spatial arrangement and the amount of leafy and woody materials separately in a forest by classification of lidar returns from a new, innovative, lidar scanner, the Dual-Wavelength Echidna® Lidar (DWEL). The DWEL uses two lasers pulsing simultaneously and coaxially at near-infrared (1064 nm) and shortwave-infrared (1548 nm) wavelengths to locate scattering targets in 3-D space, associated with their reflectance at the two wavelengths. The instrument produces 3-D bispectral "clouds" of scattering points that reveal new details of forest structure and open doors to three-dimensional mapping of biophysical and biochemical properties of forests. The three parts of this dissertation concern calibration of bispectral lidar returns; retrieval of height profiles of leafy and woody materials within a forest canopy; and virtual reconstruction of forest trees from multiple scans to estimate their aboveground woody biomass. The test area was a midlatitude forest stand within the Harvard Forest, Petersham, Massachusetts, scanned at five locations in a 1-ha site in leaf-off and leaf-on conditions in 2014. The model for radiometric calibration assigned accurate values of spectral apparent reflectance, a range-independent and instrument-independent property, to scattering points derived from the scans. The classification of leafy and woody points, using both spectral and spatial context information, achieved an overall accuracy of 79±1% and 75±2% for leaf-off and leaf-on scans, respectively. Between-scan variation in leaf profiles was larger than wood profiles in leaf-off seasons but relatively similar to wood profiles in leaf-on seasons, reflecting the changing spatial heterogeneity within the stand over seasons. A 3-D structure-fitting algorithm estimated wood volume by modeling stems and branches from point clouds of five individual trees with cylinders. The algorithm showed the least variance for leaf-off, woody-points-only data, validating the value of separating leafy and woody points to the direct biomass estimates through the structure modeling of individual trees.