Enhanced 3D Sub-Canopy Mapping via Airborne/Spaceborne Full-Waveform LiDAR
Abstract
Problem Identification: Light Detection and Ranging (LiDAR) technology has been used for decades to measure 3D vegetation structure, but it is not abundantly apparent how the laser light interacts with/propagates through the objects it encounters, and ultimately, which portions of the 3D space have been adequately sampled, or more critically, have remained unsampled. This is confounded by the fact that airborne LiDAR can measure the top canopy surface well, but typically is unable to capture detailed deep-canopy or understory structure, and sometimes does not even penetrate to the ground. This lack of sub-canopy interactions is attributed to i) attenuation of the LiDAR (waveform) throughout the canopy elements, ii) occlusion effects due to opaque structures, and iii) discrete LiDAR hardware limitations. However, while 2D coverage and point density are easily analyzed, there are currently no tools to assess how comprehensively the 3D space was sampled. We contend that analysis of full waveform LiDAR (wLiDAR), along with first principles radiative transfer modeling, will improve our ability to analyze targets and gaps, and identify portions of the beam path where remnant laser photons are too low to illuminate a target/voxel for subsequent detection. Approach: The proposed research differs from previous studies in that we will: i) use wLiDAR data, thus allowing a higher resolution analysis of all targets located along the laser beam path within each voxel; ii) perform an analysis of the horizontal beam width, including partial beam occlusion within each voxel; iii) generate advanced data products, such as fraction classification (target/empty/occlusion) and understory maps; iv) analyze plant area index and leaf area index per voxel; v) provide scene truth with detailed knowledge in terms of which output voxels are filled/empty (the simulation portion of the project); and vi) analyze the physical target cross-section vs. the resulting voxel classification, thus allowing a confidence to be assigned to the result. Anticipated Results: High-fidelity simulation truth data and field sampling will advance the analysis and confidence of output products, resulting in more 3D voxelized output classes and biophysical parameters than previous studies. Simulated waveforms, generated with Rochester Institute of Technology’s (RIT) DIRSIG software, will be used for algorithm development (PI: Dr. Jan van Aardt), after which the processing algorithms will be extended to real airborne wLiDAR data, via collaborator Battelle/NEON (National Ecological Observatory Network; co-PI: Dr. Keith Krause). Laser pulses will be analyzed and interpolated onto a voxel grid to assess the 3D distribution of targets, gaps, unclassified, and unsampled areas, and verified via simulated scene truth data. We also propose value added products for Option Years 3-5: an understory map, extensive waveform signal processing (e.g., transmission and attenuation correction, denoise, and deconvolution), and 3D segmentation of filled/empty spaces. Impact on and Contributions to NGA Capabilities: Outcomes will help to advance NGA’s ability to produce relevant and accurate GEOINT (Research Directorate Goals 1 & 2): creating new capabilities that automate characterization of objects, features, and activities of interest under forest canopies; and developing advanced GEOINT capabilities and tradecraft in LiDAR remote sensing. This effort will enable advanced assessment of wLiDAR sensor performance, improved analysis of wLiDAR for object detection, increased confidence in wLiDAR data products, and utilization of automated signal/image processing workflows.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 29, 2020
- Source ID
- HM04762010008
Entities
People
- Jan Van Aardt
Organizations
- National Geospatial-Intelligence Agency
- Rochester Institute of Technology