In recent years, light detection and ranging (LiDAR) has become prevalent in forest resource management. Specifically, Terrestrial Laser Scanning (TLS) has gained wide adoption in forest remote sensing due to its high flexibility and ability to generate high-quality point cloud. Classifying the point cloud into key forest components, such as ground, understory vegetation, stem, and foliage, is a vital step in effectively managing and comprehending forest ecosystems. However, the diversity and complexity of tree morphology make the point cloud data huge, and the irregular crown makes it difficult to classify branches and leaves. Improving segmentation accuracy and data processing efficiency is a challenging task. Traditional threshold segmentation and clustering analysis methods have great differences in dealing with trees with complex structures and point clouds from different parts of the trees. Currently, 3D point cloud segmentation network still has limitations in processing large-scale point cloud data.

In this study, we proposed an automatic method for segmenting TLS point cloud of trees using a 3D semantic segmentation network. A powerful local feature extractor was introduced through a 3D convolution method which was tailored for processing large-scale point cloud data. A global contextual integration module was also introduced for better feature fusion. Additionally, a semantic context encoding loss was introduced to supervise the network to integrate abundant semantic context features. To evaluate the network, we established a TLS dataset with precise manual labeling, containing over 50 tiles of multi-platform, multi-scale data in diverse forest environments. It covers typical forests in China and Europe. Results indicated that the network is generalized from labeled to unobserved data, which shows the characteristics of high accuracy and stability. The research presented in this paper provides a new approach for the automatic and accurate segmentation of forest point cloud for related fields of forestry and vegetation parameter extraction.

The abandonment of agricultural land in the United Kingdom has enabled the regrowth of vegetation through the passive rewilding. Unlike human-assisted afforestation programs, passive rewilding involves the colonization of tree species through dispersion by wind or animal and facilitates natural succession. Adjacent to Monks Wood National Nature Reserve, an ancient woodland, are two areas that have been undergoing passive rewilding since 1961 (Old Wilderness: 3.9 ha) and 1996 (New Wilderness: 2.1 ha). Within these areas, bird species like Chiffchaff (Phylloscopus collybita) and Willow Warbler (Phylloscopus trochilus) occupy their preferred habitat in mature and early successional forest areas respectively, the structures of which can be characterized with airborne laser scanning (ALS). The aim of this study is to quantify changes to the availability and distribution of habitat that these specialists occupy. Metrics characterizing the full vertical profile and strata representing shrub, understorey, and overstorey layers were extracted from the ALS data. A time series of ALS-derived structural metrics, along with bird survey data, were used to develop habitat models characterizing Chiffchaff and Willow Warbler habitat. Modelling was completed using Random Forests, which also ranks variable importance. Maximum height of the full vertical profile was found to be important to both bird species, with Chiffchaff occupying the higher overstorey strata and Willow Warbler occupying the lower shrub layer. Through passive rewilding, the availability and distribution of forest structures has changed in the Old and New Wilderness areas. For Chiffchaff, these changes meant that there has been a gain in the availability of its preferred habitat, to which it appears to have been able to capitalize as reflected in the increased number of individuals observed over time. Conversely, Willow Warbler habitat has diminished and fewer individuals have been recorded over time, though clusters of its preferred habitat remain.

Knowledge of multi-scale and multi-species forest chlorophyll contents is critical for research on forest ecology and carbon cycle. Hyperspectral LiDAR (HSL) technology has the advantage of simultaneous acquisitions of 3D spatial and hyperspectral information, which provides advanced technology for forest applications. However, it is unclear the ability of the HSL technology to monitor multi-scale and multi-species forest chlorophyll contents. Therefore, this study investigated the potential of HSL technology with radiative transfer models (RTMs) to estimate multi-species leaves, leaf-scale and individual-scale forest-simulated chlorophyll contents at a laboratory. Through the HSL prototype system, hyperspectral data were measured for leaf-scale samples, and hyperspectral data and 3D spatial coordinates were measured for individual-scale samples. At the leaf scale, three leaf RTMs (PROSPECT-4, PROSPECT-5 and PROSPECT-D models) were utilized to estimate chlorophyll contents of multi-species leaves. At the individual scale, coupling the best PROSPECT with optimized 4SAIL models was used to estimate and map 3D chlorophyll distribution. Results showed that the PROSPECT-5 model had the best performance (R²=0.60), followed by PROSPECT-4 (R²=0.58) and PROSPECT-D (R²=0.56) models for multi-species leaves. The best PROSPECT with optimized 4SAIL models achieved an R² of 0.66 at the individual scale. The mapped 3D chlorophyll distribution agreed with the actual chlorophyll distribution. The HSL technology has the potential to become an alternative tool for accelerating forest ecological research and monitoring.

We present a new method for converting forest lidar scans into a piecewise-cylindrical representations of the individual trees. Unlike existing reconstruction methods that segent trees first and then reconstruct the branches afterwards, we reconstruct the tree branches of all trees simultaneously. This lets tree connectivity information inform the tree segmentation, and so allows accurate tree segmentation even when branches are intertwined between trees. We demonstrate the results on hectare-scale regions of savanna woodland and temperate forests, showing both robust tree segmentation and detailed tree branch reconstruction, with results available within minutes.

We also present our approach to estimating reconstructed tree branch diameters in the presence of noise, foliage and partial lidar coverage; and we introduce Leonardo's tree branching rule as a prior in order to improve branch diameter accuracy. This rule also supports the reconstruction of dead trees and branches, which are not supported by existing tools.

Good tree reconstruction in natural environments relies on dependable ground reconstruction in order to differentiate ground structures and undergrowth from trees. We include a summary of our novel zero-parameter method of ground reconstruction that is robust to variations in point density, and is solved by employing fast Pareto front estimation from the field of multiobjective optimisation.

Finally, we present the software tools (RayCloudTools and TreeTools) that we have developed to support the full pipeline from point cloud to the manipulation and analysis of these tree files. The presented algorithms are therefore available as open source for the community to use.

In the context of its energy transition, Reunion Island aims to develop a wood-energy sector to provide local, renewable fuel to power plants and reduce reliance on imported resources. Acacia mearnsii, an invasive exotic species, is identified as the main potential source of wood-energy on the island, but it is still poorly documented and characterized. Therefore, an accurate assessment of this resource is a key step in this effort. The successive tropical storms led to complex stands with numerous inclined or entangled trees, making the resource modeling challenging.

This study aims to explore methods based on high-density LiDAR to improve the resource evaluation of Acacia mearnsii. Two sub-objectives were identified: 1) build quantitative predictive models for estimating basal area, wood volume, and biomass from LiDAR variables using an Area Based Approach (ABA), and 2) produce a map of different types of stands to locate most suitable stands for mechanized exploitation, i.e. excluding stands with entangled trees.

Fieldwork was carried out in 2022 in the “Hauts Sous le Vent” forest, at an altitude of about 1500 m in the West of the island. Allometric equations specific to Acacia mearnsii were established and density measurements were taken on a sample of wood collected to convert volume and biomass. One hundred plots were inventoried, ten of which were fully measured to establish the allometries, while UAV LiDAR data were collected.

Preliminary comparison of basal area surface models for straight stands and for all stands showed discrepancies of residual errors (see Figure). Ongoing work aims to extract tree inclination proxies from LiDAR data in order to improve models to predict stand characteristcs and contribute to stand classification. The study demonstrates the potential of high-density LiDAR data to better characterize Acacia mearnsii stands and provides valuable information for the wood-energy sector on the Reunion island.