Understanding the spatial distribution of basal area growth rates within stands is crucial for implementing precision thinning that seeks to release trees and increase overall basal area growth. High-density drone lidar enables tree-level approximation across managed forest stands that can produce fine-scale inventories of stand structure. Segmentation of these high-density point clouds allows for the quantification of tree-level metrics that drive diameter growth, such as crown size, while also describing key environmental drivers, including solar irradiance and soil moisture. We propose the capacity to quantify variability in these drivers at the tree-level can enable the fine-scale description of basal area growth rates.

High-density lidar data was acquired from the Zenmuse L1 sensor mounted on a Matrice 300 RTK drone over five mid-rotation stands in British Columbia, Canada. We stem mapped and measured 815 trees using a Haglöf Postex and matched to our lidar data using simultaneously acquired 1cm RGB imagery. Tree cores were extracted and measured from 150 stem-mapped trees and used to quantify five year basal area growth patterns. We then investigated the ability of tree-level metrics to describe observed variability in recent basal area growth. To do so we incorporated local competition through distance dependent indices derived from lidar crown metrics that describe each tree’s dominance relative to its neighbours. We found crown volumes derived from alpha-shapes and solar irradiance generated with rayshading simulations explained a large amount of variability in recent growth (R² 0.67).

This work demonstrates the capacity of a readily available lidar systems to reliably approximate and describe individual crown growing conditions which drive variations in diameter growth across dense managed stands. The estimation of tree growth patterns from lidar-derived tree form, while well demonstrated with terrestrial laser scanning; appears possible from drone platforms, providing wall-to-wall descriptions of dense forest stands critical for operational management.

Forest growth dynamics analyzes play a key role in determining carbon accumulation and balance of forest ecosystems. Understanding these dynamics brings environmental and socio-economic contributions regarding forest productivity and management. However, the specific mechanisms of how tree canopies grow, interact, and fill in the gaps in three-dimensional space remain insufficiently understood. Laser scanning systems are proven tool to map 3D crown structure and understanding forest growth dynamics. With laser scanning time series, accurate 4D monitoring (3D + time) of canopy structure (i.e., gaps and crown growth directions) is nowadays possible, opening up a novel access to answering research questions, such as how trees fill the canopy gaps depending on its neighboring trees and light availability. Moreover, it is possible to assess how environmental factors shape tree crowns of different species to increase their survival and growth rate. A study case is presented to demonstrate the use of LiDAR time-series to detect horizontal canopy growth of 50 trees in a boreal forest and how the growth relates to the gaps between their crowns. The LiDAR time-series data was collected by LiDAR phenology station (LiPhe), which consists of a Riegl VZ-2000i scanner placed in a 35-metre tower above forest canopy and tilted 60 degrees, enabling a unique 4D monitoring of forest with dense temporal and spatial resolutions. Here, we demonstrate changes in tree crown outlines of three boreal species (silver birch, Norway spruce and Scot pine) that were monitored between Apr. 2020 and Apr. 2021. We measured each tree’s crown area, relative horizontal area growth, and the azimuthal growth direction for three time points. We investigate the correlation between the growth parameter dependency and environmental factors, such as neighboring species and competitive index.

Delineation of individual trees in stands and subsequent part-level segmentation from high-density 3D laser scans are essential tasks for precision forest management and tree phenotyping research. Traditional rule-based methods such as quantitative structure models can produce good results but require manual tuning of parameters and guidance by operators – limiting their scalability and transferability. Methods based on 3D deep learning offer an alternative approach for instance and part-level segmentation using dense point clouds. These techniques have produced substantial advances in the ability to accurately classify 3D laser scans used in the fields of self-driving vehicles, indoor scene understanding and robotics. The success of these models has been driven by the availability of large and accurate labelled datasets which can be used to parameterise the weights of deep neural networks. Obtaining accurately labelled datasets from high density forest scans is extremely labour intensive and subjective, making it impractical to achieve the scale required. In this presentation we outline a new approach to generating synthetic point cloud data from high-fidelity 3D simulated scenes of forests. We describe the toolchain we have developed to generate biologically-based tree simulations to render 3D trees at different growth stages, including competition and environmental interaction before importing these simulated stands into popular 3D engines. We describe the process of generating realistic synthetic 3D scans using the popular Helios lidar simulator and a GPU-accelerated lidar simulation engine designed for robotics and self-driving applications. We evaluate the efficacy of our simulation pipeline by training deep learning instance segmentation models on our synthetic data and evaluating on both withheld synthetic data as well as hand-labelled datasets acquired for a tree phenotyping research project. Lastly, we evaluate the strengths and weaknesses of this approach by contrasting the workflow with the QSM-based workflow currently used in our phenotyping pipeline.

Forests are dynamic ecosystems under constant change, with the most natural reason for change being the tree growth. The occurrence of changes as well as factors behind and resulting from them are of interest to many. Important topics are e.g., allocation of tree growth, up to date information about forest resources, the effects of tree growth to carbon sequestration potential of forests and quality of timber.

Typically tree growth has been determined by performing repeated measurements of basic tree attributes. However, more in detail and accurate information from tree growth as well as changes in the stem form and structure of the trees is needed to be able to further understand the development of trees and forests in different conditions and their reactions to the changing environment.

Terrestrial laser scanning (TLS) allows one to characterize trees and their surrounding environment on millimeter scale. This study presents results from the use of TLS point cloud based automated tree detection and measurement methods. The results show the ability of the used method to determine changes in attributes such as diameter, height and volume of individual trees as well as allow one to successfully quantify statistically significant changes resulting from tree growth in different growing conditions both in stem form of trees and forest structures.

When investigating interactions between tree growth and its neighborhood from the point clouds, clear correlations were found between increments in stem dimensions and detailed 3D characterizations describing crown structure and competition. Observing seasonal increment in tree structures was also possible with TLS data, if a certain minimum threshold for the increase is exceeded to ensure measurement accuracy. The findings show the capability of TLS point clouds to measure tree growth and characterize changes in forest structure as well as improve understanding of the processes related to them.

Forest stratification is a powerful tool to determine the vitality of forest stands with implicit forecasting support. It enables the extraction of vital information such as biomass, forest physiology and rejuvenation capacity [1-4]. Given their importance, in situ stratification studies are performed despite their complex nature and high intrinsic costs (time, people and equipment). These studies entail per-tree manual parameter acquisition, such as crown height and width, tree height and trunk radius [5]; and statistical models [6]. In contrast, airborne laser scans (ALS) and unmanned aerial vehicle laser scans (ULS) can generate high-density wide-area scans but lack annotations crucial to derive information and train neural networks. To contribute towards automating the classification of stratification, we:

• propose a generative forestry scene pipeline that merges real and synthetic tree objects into ALS/ULS-like point clouds and provides their per-point annotation;

• study this data usage for training a point cloud segmentation neural network for stratification [7,8];

• measure the model performance on real data to assess its viability as a stratification tool.

The pipeline exploits synthetic trees and vegetation descriptor values like height and radius by randomly permuting them within defined limits, subsequently generating an arbitrary number of parametrised mesh objects. Besides, hand annotated real ULS and terrestrial laser scans are captured for training and validation. Synthetic and real objects are then randomly placed in a scene to simulate ULS data acquisition [9]. The resulting point cloud is annotated using the known relation between the object parameters and its point cloud representation.

Although the pipeline can create arbitrary trees, this contribution focuses on alpine forest stands with various species, such as larch, spruce and beech. Results verification is conducted using real ULS and terrestrial laser scans collected in Ebensee, Upper Austria. Moreover, we explore potential benefits of mixing real with synthetic annotations for training.