Integrating LiDAR Sensing, Simulation, and Decision Support for Automated Assessment of Tree Stem Quality in Cut-to-Length Forest Harvesting

Abstract

Advancements in precision forestry, sensor technology, and automation are reshaping the future of cut‑to‑length (CTL) harvesting by enabling more data‑driven and reliable decision‑making during forest operations. This dissertation investigates how LiDAR‑based sensing, simulation frameworks, and analytical workflows can be integrated to support automated detection of tree stem quality and harvesting outcomes in machine-assisted thinning operations. Traditional assessment of stem quality and harvesting performance relies heavily on operator experience, visual judgment, and labour‑intensive field surveys. These approaches are susceptible to subjectivity, operational variation, and limited spatial coverage. As thinning operations become more frequent and forest managers seek greater consistency and efficiency, there is a growing need for methods that provide objective, scalable, and real‑time insight into forest conditions, tree‑level quality, and operational impacts.

To address these challenges, this research develops and evaluates a comprehensive sensing and analysis framework that leverages both machine‑mounted and mobile laser scanning (MLS) systems, complemented by simulation‑based experimentation. Four interconnected studies form the core of this work. The first study identifies industrial needs, operator requirements, and relevant sensing technologies to detect stem defects—such as curvature, crookedness, and other geometric anomalies—that affect value recovery. The second study integrates a LiDAR sensor into a CTL harvester to investigate the feasibility of real‑time stem‑quality detection before felling. Through detailed point‑cloud analysis, the study demonstrates that onboard LiDAR can measure diameter, straightness, and other key attributes with promising accuracy, opening opportunities for semi‑automated decision support.

Recognising the high cost and logistical challenges of field testing, the third study introduces a simulation framework—the Tree Classification Framework (TCF)—for controlled development and validation of detection algorithms. By generating synthetic forests, point cloud datasets, and defect scenarios, the framework enables repeatable experimentation and algorithmic benchmarking across varying conditions. The final study extends the approach to post‑harvest evaluation by utilising backpack MLS to assess thinning quality in operational stands. Attributes such as strip‑road spacing, residual stand density, stump dimensions, and stem defects are extracted and compared with conventional field measurements, demonstrating strong agreement and highlighting MLS as a reliable method for operational quality monitoring.

Across all studies, the methods developed in this dissertation demonstrate that LiDAR‑derived point clouds—whether generated in simulation, captured on board harvesting machines, or collected via mobile scanning—provide a rich and accurate representation of tree geometry and harvesting outcomes. Data‑processing workflows incorporating segmentation, curve fitting, and classification techniques enable consistent measurement of stem attributes and objective detection of defects. Combined with statistical analysis and validation against ground truth, these techniques reveal the strengths and limitations of LiDAR sensing for forest‑machine integration and harvesting‑quality assessment.

The dissertation contributes an end‑to‑end framework for automated forest‑quality assessment that links sensing, modelling, algorithm development, and field validation. The findings underscore the operational potential of integrating LiDAR sensing into CTL harvesters and support the emerging role of MLS in post‑harvest auditing. Together, these approaches offer a pathway toward intelligent harvesting systems that provide actionable information to operators, improve thinning consistency, and enable more sustainable and efficient forest management. By bridging technological innovation with practical forestry needs, this research lays the foundation for future advances in automated decision support, autonomous machine functions, and data‑driven forest operations.