Epsilon Archive for Student Projects

Abstract

The structure of a forest ecosystem is an important ecological, environmental, and socio-economic driver. Forest remote sensing is witnessing a rapid development of new technologies and methodologies for quantifying forest structure; however, they are not without flaws. Laser scanning, as one of the major data collection methods in RS, is negatively affected by both the scanned environment and the human error. This master’s thesis aims to address and quantify two of the established drawbacks of LiDAR scanning - occlusion, and user focus bias, to further help with LS optimalisation.
A new voxel-based raytracing package for the programming language Julia was developed to analyse occlusion, user focus bias and openness of point-clouds. It merges the start and endpoints of laser beams to construct laser rays, which traverse the voxel environment. In contrast to traditional point cloud analysis, which considers only the endpoints of the beams, introducing ambiguity into derived metrics and losing parts of the scanned information, our raytracing method provides a more comprehensive approach. The introduced raytracing method was demonstrated on the P-TLS platform. With the raytracing analysis performed on five demonstrational scans, we were able to illustrate the effect of user focus bias on occlusion and derived tree height. Additionally, we proposed a new method of measuring tree height using the openness quantifier – the G-T method. In our analysis, the traditional method of deriving tree height using a percentile of a quantile (P90 and P95) greatly underestimated the presumed height, as well as showed susceptibility to the user focus bias quantifier, unlike the more robust G-T method.