The development of close-range technologies avails us of enormous data for assessing forest inventory parameters. Forest inventory is required for proper forest management as it provides statistical support and deals with different aspects of challenges in forestry. In this sense, some instruments give precise and accurate information in forest inventory, such as TLS (Terrestrial Laser Scanner) and MLS (Mobile Laser Scanner). However, a proper data collection methodology to implement them for forest inventory approaches has yet to be available. So, there is a need to develop a methodology for data acquisition to get more insight into the forest.

This research aims to develop a methodology suitable for data acquisition and optimization of plot design of different tree parameters using MLS. Later, this data and control field data will be used for installing and testing the algorithms and standalone software using the point cloud data to find the best processing solution for forest inventory.

Documentation will be available at the end, including the solution of all the possible point cloud processing algorithms and standalone software. This may also include the latest or new version of the software or tools. These algorithms and software will be based on programming languages such as Python, R language, C++, Matlab, Java, etc. A detailed list will be prepared for all the possible and publicly available software and algorithms for the forest PCD (point cloud data). The performance will be assessed using tree parameters used in forest inventories, such as tree height, DBH (Diameter at Breast Height), stem volume, tree detection, etc.

The main output will include the possible methodological scheme instructions and solutions describing the data collection and processing by software and algorithms for the end users in forestry inventory. This contribution is supported and based on Reforest Project no: 101060635.

Forest inventory and management practices are necessary but labor-intensive and expensive using conventional ground sampling methods. The demand for an automated and cost-effective solution is increasing among stakeholders. Mobile LiDAR systems, benefiting from simultaneous localization and mapping technology, have demonstrated advanced capabilities in navigating within forest canopies and determining individual tree attributes such as diameter at breast height (DBH) and stem volume in relatively open or plantation forests. However, few studies have validated the effectiveness of this technology across a variety of forest conditions, particularly in Canada, where forest dynamics are significantly diverse. This study aimed to investigate the usability and accuracy of mobile LiDAR data for automatically deriving basic tree attributes at the plot level in variable forest conditions.

LiDAR data was collected using the GeoSLAM LiDAR system over 16 plots distributed across four boreal and mixedwood forest sites in Canada, including different species, age, tree density, and terrain conditions. Single-tree stem location, DBH, and total height were derived from an automatic workflow developed within the Computree platform and compared with field measurements. The results showed that height estimation from mobile LiDAR was consistent across sites with an average of about 5% bias for the leading softwood species, meeting operational standards. The average absolute variation of DBH varied across sites from about 5% to 12%, depending on various factors. Omission and commission errors in stem detection were primarily impacted by data acquisition quality and understory complexity. Tuning of noise filtering parameters can lead to higher accuracy in tree detection and DBH estimation.

The cross-site examination of mobile LiDAR for operational forest inventory revealed its technical advantages and practical challenges from various perspectives. This study contributes to the understanding of the effectiveness of mobile LiDAR in different forest conditions and provides guidance for optimizing parameters to improve accuracy.

Low-cost Colour and Depth (RGB-D) sensors are seeing increased integration into consumer devices, such as phones and tablets. These sensors have been shown to characterise common basal inventory measurements, such as stem position and Diameter at Breast Height (DBH), across urban, plantation, and native forest environments. As such, consumer RGB-D devices provide the opportunity to increase the accessibility of proximal 3D remote sensing for forest management, both in terms of hardware cost and familiarity of use. However, one of the major limitations of active projection RGB-D sensors is that their range of measurement is limited, capturing depth information up to a range of <6m under optimal conditions. This limits their ability to accurately capture key inventory metrics representing vertical structure such as sweep, taper and tree height. Furthermore, Simultaneous Localisation and Mapping (SLAM) algorithms, used to register depth frames and provide near-real time point cloud representations to operators, are subject to positional drift, the error of which increases with capture time.

Structure from Motion (SfM) photogrammetry, captured from consumer drones, suffers from the inverse, providing representations of canopy structures but missing understory structures beneath occluding canopy. By fusing these two low-cost datasets, partial representations of vertical structure visible through canopy gaps may improve RGB-D inventory measurements whilst attributing terrestrially captured structure to canopy information.

This presentation explores the benefits obtained through the co-registration of these two low-cost datasets when representing a 30m by 50m plot of native south-east Australian eucalypt forest. It also describes the accuracy of this fused approach, compared to a Terrestrial Laser Scanning (TLS) validation dataset, when acquiring common inventory metrics of stem DBH, location, and height to canopy; and how well vertical forest structure is represented.

Canopy structure is an emergent property of forest ecosystem that reflects the spatial organization of tree trunks, branches, and leaves within the canopy. It is a key attribute for predicting the ability of forests to store carbon and provide habitat for biodiversity. With growing access to remote sensing technologies such as airborne LiDAR, characterizing canopy structural attributes at scale has become increasingly widespread in ecology. However, it remains unclear how robust different types of LiDAR-derived canopy structural metrics are when comparing data acquired using different platforms and flight specifications. To address this knowledge gap, we collected LiDAR data at 117 plots (4-ha each) in Australia’s Great Western Woodlands using REIGL sensors mounted on both unoccupied aerial vehicles (ULS) and airplanes (ALS). Using these data, we calculated 33 metrics describing canopy structure that have been previously proposed in the literature, including ones relating to canopy height, openness, and vertical complexity. These included a combination of metrics quantified directly from the LiDAR point clouds and from derived canopy height models (CHM), enabling us to compare the robustness of both types of metrics across the two platforms. We found that metrics relating to canopy height and ones capturing properties of the outer surface of the canopy tended to be highly consistent between sensors (low bias and high precision), particularly when derived from CHMs. By contrast, metrics deigned to capture the arrangement, volume and complexity of vegetation in vertical space often varied considerably between ULS and ALS datasets (high bias and low precision), especially when derived from point cloud data. Our study takes an important step towards identifying platform-agnostic canopy structural metrics that can be used with confidence for ecological studies that rely on LiDAR data acquired using a range of instruments and platforms.

Canopy structural complexity (CSC) measures derived from terrestrial light detection and ranging (LiDAR) data are strongly correlated with forest carbon cycling processes at stand to landscape scales. In this study, we propose methods for processing LiDAR data collected by a small-footprint airborne platform to derive three metrics that are strong drivers of forest Net Primary Production – Canopy Rugosity, Top Rugosity, and Rumple. Robust estimation of CSC metrics using airborne laser scanning (ALS) data is essential to scaling and modeling ecological processes at large spatial scales. However, the estimation of these metrics can be highly sensitive to instrument and collection parameters. Due to hardware limitations and logistical restrictions, current ALS systems are restricted in terms of coverage, duration, repeat flights, and flight parameter configurations. Therefore, we cannot investigate the impact of using various collection settings on the derivation of such structural products. We thus used DIRSIG (Digital Imaging and Remote Sensing Image Generation) , a physics-driven simulation environment for generating high-fidelity remote sensing data, to simulate various ALS collection settings. Specifically, we simulated multiple flights of NEON’s (National Ecological Observatory Network) AOP (Airborne Observatory Platform) on a spectrally and structurally accurate virtual recreation of a 500mx700m scene of Harvard Forest, Massachusetts, USA. Simulated datasets varied in terms of LiDAR footprint size, pulse density, and outgoing pulse width – which respectively determine the sampling area, sampling density, and range resolution. Using the exact leaf area in the voxelated DIRSIG scene, we computed precise ground truth of each metric, coincident with the simulation transects. We then performed linear correlation analysis to investigate the effects of the selected collection parameters on the estimation of each metric. Based on our results, we will recommend the optimal flight configuration for robustly estimating each metric.