Imaging Science MS Thesis Defense
A Step Towards Developing Digital Twins of Forest Environments with Terrestrial Laser Scanning
Zoom Link here
Aisha Co
Imaging Science MS Candidate
Rochester Institute of Technology
Abstract:
Digital twins have become an increasingly relevant topic as we pursue accurate representations of the real world for planning, research, and even design purposes. This study focuses on the steps for creating digital twins of forest environments using sparse points from terrestrial light detection and ranging (LiDAR). Specifically, we developed geometric models of structural digital twins (SDTs) from leaf-off terrestrial laser scanning of real-world forests. Three algorithms were used to develop an approach to handle individual tree reconstructions for sparse point clouds, and they were tested and validated with a two-sample Kolmogorov-Smirnov statistic against an existing inventory of the RIT Nature Trail to evaluate the reconstruction’s fidelity. The key evaluation metrics included diameter at breast height (DBH) and height. The limited nature of available ground truth necessitated the use of an additional validation method using a PANGAEA open dataset, based on comparing DBH, height, and crown diameter against the dataset’s calculated measurements. Reconstructions were compared against per-tree reference values at two-point spacings for sixteen trees from the open BR07 dataset (PANGAEA); these specifications matched the collected RIT Nature Trail data and the scanner’s nominal spacing. On BR07, the proposed method had the lowest median DBH error among the methods that modeled every tree (1.28cm, 4.1%), and this error was nearly unchanged at the coarser spacing, while TreeQSM’s rose from 4.2% to 6.5% and AdTree overestimated DBH by 30% to 40%. Results show the proposed method understates the tree heights by 0.2m to 1.4m but contains the least error when estimating DBH. The proposed method was able to produce all 69 trees, while SimpleTree produced 49 and none at the 4.35 cm BR07 spacing. We concluded that the developed geometric model (digital) twins have significant potential to serve as both simulation and prediction tools. We recommend that future efforts focus on collecting per-tree ground truth measurements within the scanned plot and incorporating leaf-on scans. Leaf-on scans include representative foliage, which would add realism to the model and ecological function of the trees.
Intended Audience:
Beginners, undergraduates, graduates. Those with interest in the topic.
To request an interpreter, please visit myaccess.rit.edu
Source: www.rit.edu
