Bridging the scale gap : drone‑based lidar for assessing functional forest diversity

<p dir="ltr">Given the increasing destruction of habitats and the effects of climate change, it is essential to understand and measure biodiversity for successful conservation and management of ecosystems. While species richness has traditionally been the metric of choice, Functional Diversity (FD) — encompasses the measurement of the range, abundance, and distribution of functional traits within an ecosystem — offers a more nuanced view of ecosystem resilience and function. The unique ability of Functional Diversity (FD) to be applied to multiple ecological scales, from genes to ecosystems, makes it well suited to be combined with remote sensing data. Remote sensing technology has revolutionised our ability to map and monitor FD at unprecedented scales. However, there exists a scale gap between the fine-scale measurements available through fieldwork and the broader scope of satellite observations. This thesis leverages a multi-scale approach, capitalising on the potential of drone remote sensing technology, to bridge this gap. Drone-based methods offer a promising avenue to capture FD at scales that are both ecologically meaningful and that facilitate effective conservation actions. The physical structure of trees is a fundamental element of any forest ecosystem, playing a key role in providing habitat for other species, as well as in climate regulation, water balance, and human welfare. By measuring tree structure, insight can be gained into productivity and biomass, which are essential for assessing biodiversity. Drone- or UAS-based Laser Scanning (ULS) offers a unique opportunity to collect structural data from individual trees at the operational, intermediate scale between ground observations and satellite data. This bridges the gap between ecological and operational scale, thereby advancing our understanding of individual tree functions and their interactions. Deriving functional forest diversity from ULS data is a promising approach to linking ground measurements to large-scale satelitte-based observations to monitor forest biodiversity. However, the efficacy of high-point density ULS data in the context of FD assessments has yet to be demonstrated. Furthermore, the high point-density provides the opportunity to derive new insight into the drivers of individual tree functions. The aim of this thesis is to evaluate FD within various Australian forested ecosystems using morphological traits extracted from high-density point clouds to fill the spatial scale gap between ground observations and satellite data. The thesis comprises three core chapters, following a publication-based format. The data for this thesis is mainly derived from ULS but also includes Terrestrial Laser Scanning (TLS) data to quantify the variability in ULS derive morphological traits. Four of the six study sites are on the east coast of Australia, and the remaining two are located in Tasmania. Eucalypt trees dominate 80 % of Australian forests, constituting the ecosystem across all sites and showcasing diverse structures and species, from dense rainforests to natural monocultures. Point clouds of three sites have been segmented to identify individual trees and separate overstory from understory. Chapter 2 focuses on the use of remotely detected morphological traits to evaluate the FD of the forest. High-density lidar data from ground and drone sources allow consideration of ecologically significant traits at fine-scale units like individual trees and sub-canopy elements. FD is calculated using an Trait Probability Density (TPD) approach, which demands substantial computational intensity in high-dimensional trait spaces. Reducing dimensions via trait selection and Principal Component Analysis (PCA) alleviated the computational burden, aiding in identifying meaningful traits and minimising inter-trait correlation. I explored whether Kernel Density Estimator (KDE) or one-class Support Vector Machine (SVM) offers enhanced computational efficiency in TPD computation. Four traits — Crown Height, Effective Number of Layers, Plant–Ground Ratio, and Box Dimensions — were chosen for the TPD input. Simulations reveal that KDE performs better than SVM for high-dimensional trait spaces with numerous input traits. For five or more traits, it is advisable to employ dimension reduction techniques such as PCA. In addition, the appropriate kernel sizes, aligned with ecological target units and trait numbers, are crucial. Thus, 3–5 traits require a minimum 7 × 7 pixels kernal size. This study improves the quality of TPD computation based on traits derived from remote sensing, offering practical recommendations to improve biodiversity assessment. Chapter 3 focuses on local competition between trees in natural forests and their consequential morphological adaptations. The structural organisation and variability of forests are mainly influenced by tree interactions and resource competition. Although competition is extensively studied in plantations, its impact in natural forests remains less explored. I evaluated the impact of competition on tree structure and performance by extracting functional and competition traits from individual trees segmented in high-density ULS data. Characteristic correlation analyses were performed at three eucalyptus-dominated study sites in Australia. Our findings reveal correlations between functional traits (Crown Height, Canopy Area, Canopy Radius, Plant Area Index, Volume) and competition traits (Density, Visibility, surrounding Tree Height). Variations between sites can be related to the type of forest and the variety of species present. The most pronounced relationship is between Tree Density Index and Crown Height and Radius. Although tree density is one of the oldest and most thoroughly documented competition metrics in forest plantations, research in natural forests is limited. Chapter 4 addresses the effective quantification of landscape variations and driver identification through remote sensing-based FD assessments. Beta diversity is a measure of the differences in FD between two hypervolumes in a high-dimensional trait space. It is determined by calculating the overlap or dissimilarity between the two. There is a disparity between theoretical demonstrations of FD assessments and the availability of practical tools for implementation. The extraction and comparison of FD from high-density lidar data obtained from drones in eucalyptus forests in Australia and Tasmania is illustrated. TPD is calculated from ecologically meaningful traits, investigating the influence of normalisation on Functional β Diversity. The importance of Functional Richness, Annual Minimum Temperature, soil pH, Topographic Wetness Index, and Elevation is demonstrated as drivers of Functional β Diversity. This highlights their role in the observation of forest biodiversity over time and space. Although the available tools allow for the assessment of FD, they are not designed for geospatial workflows, and they lack the capability to produce maps of FD, which is key for monitoring FD from remote sensing data. Across the three core thesis chapters, FD is quantified through the TPD method and high-density lidar data. In Chapter 2, I adapt the TPD approach to drone data and offer insights into selecting ecologically meaningful traits and TPD computations in the context of remote sensing. In Chapter 3, I explore the potential of individually segmented trees from high-density point clouds and novel ecological traits for the assessment of structural diversity. This chapter focuses on identifying ecological drivers of structural adaptation, while Chapter 4 focuses on identifying abiotic factors. These abiotic factors are identified by assessing the Functional β Diversity between the study sites and correlating the difference in abiotic factors between the study sites with beta diversity. Furthermore, this chapter serves as a demonstration of the currently available tools for the assessment of FD with remote sensing data. In the final chapter, I review and summaries my findings and discuss potential future applications, such as developing multiscale stand-alone workflows for biodiversity monitoring. This thesis is designed to offer value not only to the remote sensing community but also to ecologists, guiding the thoughtful use of TPD from remote sensing data. Enhanced accessibility and insight into the ecological importance of high-density lidar data are envisioned outcomes. Furthermore, the lessons extracted could extend to diverse applications such as temporal monitoring, biodiversity hotspot detection, and predicting climate-induced forest morphological changes. Given climate change and human impact on ecosystems, the analysis framework and practical tools developed in this thesis have the potential to improve our ability to intricately map and understand ecosystem responses to these widespread challenges. Prospective studies could demonstrate the usefulness of TPD in monitoring with multiple temporal datasets, investigate the additional benefit of competition characteristics in evaluating FD, and develop specialised software for mapping FD.</p>