Laser Scanning represents a new opportunity in forestry to effect meaningful change in the methods used for collection and analysis of inventory data. With the ability to collect spatial information in 360 degrees, while fluidly tracking the sensor and building the final point cloud, large tracks of terrestrial LiDAR based measurements are now feasible. With the advancement in collection technology, advancement in analysis methods are necessary to utilize the surplus of information. This project investigates efficient methods for automated recognition and mapping of conifer structure within terrestrial based LiDAR point clouds.
Chapter 2 develops a methodology for using terrestrial laser scanning to show that accurate predictive lower bole volumes can be obtained. This leads to more accurate taper equations which are crucial for the estimated stand volume of all trees. Advances in remote sensing technology such as LiDAR and computer vision are quickly approaching the point where semi-autonomous, or limited human interaction, inspection and modeling of tree structure using computers is possible. Model construction estimates were analyzed alongside the Hann 2011 curves. Estimates were shown to improve on the Hann equation using the Mean of Group Mean Equation and Logarithmic Function. Model performance improved the MSE, and showed the lowest bias of the equations tested. While total tree bole estimation was not possible, the results show that improvement of bole predicted diameters is certainly possible.
Chapter 3 examined the potential for laser scanning to begin to automate log quality analysis in the field using a developed branch detection algorithm. Branch knots are an indicator tied to log value and are often difficult to properly estimate. Laser scanning offers an alternative to field estimation using Mobile Laser Scanning and computer vision. The developed algorithm performed very well identifying 94.4% of branches in the lower bole but was unable to extend further due to visual obstructions beyond the lower bole. While this paper focused on branch detection, there are many ancillary products from this workflow. Tree geolocation allows for mapping of spatial relationships, permits for better planning not only at harvest time, but also in silvicultural decisions.
The concepts discussed in this thesis serve as a starting point for future work, forming the initial framework of computationally inexpensive, information rich processes for the semi-autonomous processing of LiDAR point clouds. Future research is needed into each of the methods to determine the viability of each over a landscape of varying tree forms and species.