Using multi-platform LiDAR to assess vegetation structure for woodland forest fauna

Abstract

Abstract Vegetation structure can support biodiversity by creating a variety of microclimates and microhabitats that contribute to food and shelter for different species. For this reason, biodiversity and wildlife habitat assessments often require accurate measurements of vegetation structure. Traditional methods for measuring the three-dimensional distribution of vegetation are time-consuming and often limited to small areas or a subset of the landscape. Light detection and ranging (LiDAR) is an alternative method for collecting three dimensional information on vegetation structure and other landscape features across wide areas. For the first time, we used multi-platform LiDAR data from a terrestrial sensor (TLS) and an unmanned aerial vehicle (ULS) to investigate the relationship between vegetation structure and the diversity and abundance of birds, reptiles and amphibians in a critically endangered grassy woodland ecosystem. The first Chapter of this thesis involves TLS and ULS data collection methods, post-processing steps and exploratory data analysis. We calculated a number of variables to characterise the three-dimensional structure of vegetation across four structurally different, one hectare sites (high-tree/high-shrub, high-tree/low-shrub, low-tree/high-shrub, and low-tree/low-shrub) and compared the values of the TLS and ULS derived variables. Generally, TLS outperformed ULS by producing higher volumetric and height diversity indices within our landscape. In the Second Chapter, the relationship between TLS and ULS derived vegetation structural variables and overall bird abundance, species richness and diversity were investigated using mixed effects regression models. Models showed strong significant associations between vegetation structural variables including canopy roughness, vegetation volume, vertical complexity and the abundance of individual species and guilds. The best performing models were for individual bird species and guilds, whereas overall diversity and abundance showed less association to LiDAR-derived vegetation structural metrics. TLS and ULS based models performed similarly when identifying vegetation structural associations with bird communities and individual species. In the Third Chapter, coarse woody debris (CWD) from TLS, ULS and the combination of both datasets (Fusion) was extracted. Several topographic variables were calculated as raster imagery from LiDAR point clouds and Random Forest (RF) machine learning algorithms were then utilised to classify CWD. Noise reduction algorithms were applied to reduce noise from the classified imagery. Digital height model (DHM), surface roughness and topographic position index were important variables in classifying CWD with RF. Classification accuracy varied depending on the amount of ground vegetation cover. The impacts of ground vegetation cover on CWD accuracy in a grassy woodland were quantified and discussed. The Fourth Chapter explores the relationship between LiDAR derived vegetation structural metrics and the presence and abundance of reptiles and amphibians. Our models demonstrate that woodland reptile and amphibian populations were significantly associated with a number of vegetation structural characteristics from the selected variables, the most common of which were mean canopy height, canopy skewedness, vertical complexity, volume of vegetation and CWD. Notable relationships between herpetofauna population data and vegetation structural metrics are discussed with reference to existing literature on habitat associations for these animals. We also explore reasons why significant associations between LiDAR derived vegetation structural metrics and animal population data were not consistent across sensors and suggest directions for future research.