Remote Monitoring of Grassland Function and Services: Exploring the Prospect of Structure from Motion Photogrammetry for Characterising Habitat Quality Indicators

Abstract

Structure from motion and multi-view Stereo (SfM+MVS) is a technique for creating land surface models from optical remote sensing images. SfM+MVS is often coupled with aerial drones and has advanced considerably within spatial ecology for assessing: landslide progression, hydrological pathways, coastal erosion, arable crop yields and woody vegetation. However, there are few examples of its application to understanding grassland functions and services. Grasslands are important as they deliver a range of functions and services, including: biodiversity provision, regulation of hydrology and food and fibre provision. This thesis advances understanding of the spatial and temporal uncertainty of drone and SfM+MVS-based workflows in grassland ecology. It details new solutions which address key limitations of SfM+MVS-based workflows, and explores novel applications within conservation management in grassy habitats. A combination of traditional agronomic techniques, airborne Light Detection and Ranging (LiDAR) and drone and SfM+MVS-coupled approaches were used to develop a greater understanding of the capabilities of current techniques and technologies. This thesis developed and evaluated novel methods to advance quantitative understanding of spatial variation in vegetation canopy characteristics and biodiversity provision across temperate grasslands. The research presented herein clearly details how traditional agronomic techniques alongside a drone, SfM+MVS and LiDAR-coupled workflow can support grassland conservation monitoring schemes through resource efficient measurements of key sward canopy characteristics. This thesis extends the understanding of the capabilities, limitations and confidence in drone-based SfM+MVS for understanding grassland functions and services, aiding sound and timely management of protected features. This was achieved through answering four research questions: i) How does the choice of SfM+MVS software (including settings and user experience/time) impact spatial and temporal uncertainty of sward height measurements? ii) Can drone data capture sward height variability, and are drone and SfM+MVS-based measurements reproducible across replicate image datasets? iii) Can airborne LiDAR-derived data products be used to address the lack of sufficiently accurate and fine-grain bare-Earth elevation reference data required for generating sward height measurements from a drone and SfM+MVS-based workflow? iv) Can a drone and SfM+MVS-based workflow be used to measure key habitat quality indicators for observed patterns in nectar feeding insects in high nature value grassland habitats? Summaries of how these four research questions were addressed: I Chapter 3 details the need to consider choice of software in SfM+MVS studies. Up until now, most studies carried out employing an SfM+MVS workflow were not statistically reproducible. When designing a drone and SfM+MVS-based study it is crucial to consider differences between software and how robust the workflow, including software, are by considering the variation in the SfM+MVS-derived vegetation canopy height measurements between replicate image datasets. To address the latter point I proposed that an SfM+MVS workflow for time series analyses should capture at least one replicate image dataset. This would, at a small cost, improve the reproducibility of the results, which is crucial when monitoring fine-grained indicators of environmental change over time. The findings presented in this part of the thesis have important implications for the application of SfM+MVS in ecology as well as in other fields of Earth and environmental science. II The extent to which aerial photographs taken from a drone could deliver new insights into the spatial heterogeneity of an intensively managed grassland field were determined in Chapter 4. It is argued that fine-grained monitoring of temperate grasslands at management relevant extents is either technically or practically not possible with traditional manual approaches. Thus, an easily applied workflow which can support decision-making of grassland farmers and conservation managers, allowing the optimisation of sward management for production and/or biodiversity aims is demonstrated. The need for this workflow is evidenced in scientific literature on the conservation of grassland invertebrates and bird communities which reveals the intrinsic challenge of accurately and precisely quantifying grassland habitats, let alone at the temporal resolution required to capture the complex nature of population dynamics. III The extent to which the quality of information derived from LiDAR-based data products could be improved, and the extent to which LiDAR-based data products can be used alongside drone-based aerial photographic data to deliver new insights into the spatial heterogeneity of short-sward grassland habitats were determined in Chapter 5. It is argued that not only can a workflow centred around LiDAR-derived data products deliver accurate ground height measurements, it also addresses concerns pertaining to the lack of sufficiently accurate and fine-grain bare-Earth elevation reference data. The proposed workflow extends previous attempts at calibrating LiDAR-derived height measurements into pastures and meadows. IV Chapter 6 is set on the backdrop of the rapid decline in key grassland habitats experienced around large parts of the globe since the early 1900s which stress the importance of assessing the wider implications and often complex dynamics of change. While in-situ species population counts have and will continue to be an important asset for conservation management schemes and policy making, they are time consuming. Instead, spatial and temporal assessment of habitat quality offers resource efficient means of tracking patterns and trends in key habitat quality indicators. However, current approaches are time consuming and often fail to capture patterns of important features. Hence, in Chapter 6 a novel, resource efficient image- and LiDAR and machine learning-coupled workflow capable of delivering structural measurements at the grain size and over the extent required in conservation management was developed. Thus, there is arguably now a strong case for re-evaluating existing conservation monitoring schemes in the light of new technologies and techniques.