Characterization of Mediterranean forest formations using ground-based LiDAR systems

Forests play important ecological, economic, and social roles, contributing to carbon sequestration, biodiversity conservation, and sustainable resource management. Accurate monitoring and modeling of forest structure is critical to improving our understanding of ecosystem dynamics and guiding forest management decisions. Recent advances in remote and proximal sensing technologies, particularly LiDAR (Light Detection and Ranging), have revolutionized forest inventory techniques. Among these, Terrestrial Laser Scanning (TLS) and Mobile Laser Scanning (MLS) have emerged as key tools for acquiring detailed three-dimensional data on forest canopy, stem, and understory vegetation. These technologies provide non-invasive, highly accurate measurements of forest parameters. One major challenge in forestry applications of LiDAR is the accurate separation of woody material (trunks, branches) from non-woody components (leaves, shrubs, and ground vegetation). This distinction is essential for estimating key forest metrics, including structural attributes and biomass distribution. Over the past two decades, various methodologies have been developed to address this issue. This dissertation presents a workflow for analyzing TLS and MLS point clouds, with a specific focus on accurately separating woody and non-woody components. The proposed methodology integrates voxelization, ground filtering, and clustering algorithms, providing a streamlined approach for processing LiDAR data. The methodology was tested in Mediterranean forest contexts in Sardinia, Italy. Data collection involved both fixed-station TLS devices and handheld MLS devices. These instruments were evaluated based on their ability to capture canopy structure and understory complexity. Results demonstrated that the workflow achieved high classification accuracy. Fixed-station TLS instruments outperformed handheld devices in dense and complex forest environments due to their higher resolution and data quality. However, the MLS proved effective in simpler forest settings, offering the advantage of rapid data acquisition. An important aspect of the study was the development of a software package for the "R" statistical analysis platform, called Point cloud Interactive Computation (PiC), designed to automate the proposed workflow. The PiC package processes raw LiDAR point clouds step-by-step, generating outputs at each stage, including voxelized data, ground-separated layers, and classified woody/non-woody components. The software was also tested on two forest plots. The results confirmed the versatility and ease of use of the package, which efficiently and consistently processed data from a TLS and MLS, regardless of the type of hardware used to run the software. Despite the promising results, challenges remain in applying the methodology to highly heterogeneous forests, such as those with diverse tree species, complex crown geometries, and dense understory vegetation. Future research should aim to improve the robustness of the workflow by incorporating additional features. Furthermore, testing this methodology in non-Mediterranean ecosystems would provide insights into its broader applicability and limitations, adapting it to account for variations in canopy architecture, tree species composition, and understory density. This work underscores the potential of integrating TLS and MLS technologies with advanced analytical workflows in forest monitoring. The insights gained from this research can contribute to more effective forest management practices and enhanced ecological modeling in a wide range of forest environments

Accurate forest structure monitoring is key to understanding ecosystem dynamics and informing management. Advances in remote and proximal sensing, especially LiDAR (Light Detection and Ranging), have transformed forest inventories. Terrestrial Laser Scanning (TLS) and Mobile Laser Scanning (MLS) now enable precise, non-invasive 3D capture of canopy, stem, and understory features. A significant challenge in using LiDAR for forestry is distinguishing woody from non-woody elements (leaves, shrubs, ground vegetation), which is essential for estimating structure and biomass. Over the past two decades, various methods have aimed to address this. This dissertation introduces a workflow for analyzing TLS and MLS point clouds, focusing on the separation of woody and non-woody components. The method combines voxelization, ground filtering, and clustering algorithms into a streamlined process. Trials were conducted in Mediterranean forests in Sardinia, Italy, using both TLS and MLS devices. These tools were assessed for their ability to capture vertical structure and understory detail. Results showed high classification accuracy. TLS units performed better in dense forests due to superior resolution, while MLS devices were effective in simpler contexts, offering faster data collection. A key contribution is the development of an R software package, Point cloud Interactive Computation (PiC), which automates the workflow. PiC processes raw LiDAR data through each stage, producing voxelized outputs, filtered ground layers, and classified components. It was tested on two forest plots, showing reliable performance across different sensors and computing setups. Despite promising results, challenges remain in complex forests with varied species, crown shapes, or dense understory. Future work should enhance the workflow's adaptability and test it in nonMediterranean ecosystems to assess broader applicability. This research highlights the value of integrating TLS and MLS with analytical tools for forest monitoring, offering insights to support sustainable forest management and ecological modeling