{"id":{"repo_id":"cambridge","oai_identifier":"oai:www.repository.cam.ac.uk:1810/394756"},"canonical_url":"https://search.dev.ndltd.org/etd/cambridge/oai:www.repository.cam.ac.uk:1810/394756","repository":{"repo_id":"cambridge","name":"Cambridge University","base_url":"https://api.repository.cam.ac.uk/server/oai/request"},"display":{"title":"Fidelity versus Scalability: Deep Learning in Forest Remote Sensing","abstract":"Climate change drives complex responses in forest ecosystems, with individual trees responding differently depending on species, size, and local competition. These individual responses to stressors such as drought and temperature can drastically alter forest composition, and cannot be captured through forest-scale assessments. Monitoring forest dynamics at an individual tree scale is therefore essential for accurately assessing ecosystem responses to climate change. Traditional field inventory based on manual measurements is able to provide data at this granularity, but limits the spatial scale and temporal resolution at which forest structure and diversity can be assessed. Aerial data appear to offer a promising solution to this problem by enabling rapid capture of forest data at large scales. However, while deep learning methods have been widely used to process these data, current approaches often rely on labels drawn by eye or other subjective reference data, undermining confidence in the reported accuracy of downstream tasks. In this work, I begin to address the validation gap in deep learning-based forest monitoring methods. By developing validation approaches using high-fidelity reference data, and improving methods to automate the generation of such validation data, my thesis establishes a foundation for reliable validation of airborne or satellite-based forest monitoring. I summarise my contributions as follows: I develop a scalable, airborne approach to measuring tree crown dieback and show that when measured end-to-end, the results correlate with ground-based measurements. I use ultra high-fidelity terrestrial LiDAR to derive ground truth annotations for a common intermediate step in such approaches -- aerial instance segmentation -- and use these labels to show that accuracy is falsely inflated when measured using hand-drawn annotations. I develop a new approach based on self-supervised learning to the semantic segmentation of terrestrial LiDAR, and demonstrate that it massively reduces the need for hand-labelling - a major bottleneck to the scaling of contribution 2.","abstract_html":"Climate change drives complex responses in forest ecosystems, with individual trees responding differently depending on species, size, and local competition. These individual responses to stressors such as drought and temperature can drastically alter forest composition, and cannot be captured through forest-scale assessments. Monitoring forest dynamics at an individual tree scale is therefore essential for accurately assessing ecosystem responses to climate change. Traditional field inventory based on manual measurements is able to provide data at this granularity, but limits the spatial scale and temporal resolution at which forest structure and diversity can be assessed. Aerial data appear to offer a promising solution to this problem by enabling rapid capture of forest data at large scales. However, while deep learning methods have been widely used to process these data, current approaches often rely on labels drawn by eye or other subjective reference data, undermining confidence in the reported accuracy of downstream tasks. In this work, I begin to address the validation gap in deep learning-based forest monitoring methods. By developing validation approaches using high-fidelity reference data, and improving methods to automate the generation of such validation data, my thesis establishes a foundation for reliable validation of airborne or satellite-based forest monitoring. I summarise my contributions as follows: I develop a scalable, airborne approach to measuring tree crown dieback and show that when measured end-to-end, the results correlate with ground-based measurements. I use ultra high-fidelity terrestrial LiDAR to derive ground truth annotations for a common intermediate step in such approaches -- aerial instance segmentation -- and use these labels to show that accuracy is falsely inflated when measured using hand-drawn annotations. I develop a new approach based on self-supervised learning to the semantic segmentation of terrestrial LiDAR, and demonstrate that it massively reduces the need for hand-labelling - a major bottleneck to the scaling of contribution 2.","abstract_has_math":false,"creators":["Allen, Matthew"],"institution":"University of Cambridge","degree_name":"Doctor of Philosophy (PhD)","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Lines, Emily","Grieve, Stuart"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-07-06","date_published":"2025-07-06","updated_at":"2026-07-22T22:24:18Z","subjects":["Forest Ecology","Climate Change","Deep Learning","Machine Learning","LiDAR","UAV","Airborne","Computer Vision"],"languages":["eng"],"rights":[],"rights_urls":["https://www.repository.cam.ac.uk/bitstreams/398cd768-7262-4888-afd3-4617fdb64a1d/download","http://purl.org/NET/rdflicense/allrightsreserved"],"identifier_entries":[]},"links":{"outbound_url":"https://doi.org/10.17863/CAM.124542","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Lines, Emily","Grieve, Stuart"]},{"key":"dc:contributor.sponsor","label":"Sponsor","values":["Cambridge Philosophical Society Trinity Hall"]},{"key":"dc:creator","label":"Author","values":["Allen, Matthew"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2025-07-06"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Cambridge"]},{"key":"dc:relation.isreferencedby.uri","label":"Dc Relation Isreferencedby URI","values":["https://www.repository.cam.ac.uk/handle/1810/394756"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["Doctoral"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Forest Ecology","Climate Change","Deep Learning","Machine Learning","LiDAR","UAV","Airborne","Computer Vision"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["https://www.repository.cam.ac.uk/bitstreams/398cd768-7262-4888-afd3-4617fdb64a1d/download","http://purl.org/NET/rdflicense/allrightsreserved"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.17863/CAM.124542"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://www.repository.cam.ac.uk/bitstreams/b22c7a5f-6b21-4927-933a-5bb21ca24b15/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Climate change drives complex responses in forest ecosystems, with individual trees responding differently depending on species, size, and local competition. These individual responses to stressors such as drought and temperature can drastically alter forest composition, and cannot be captured through forest-scale assessments. Monitoring forest dynamics at an individual tree scale is therefore essential for accurately assessing ecosystem responses to climate change. Traditional field inventory based on manual measurements is able to provide data at this granularity, but limits the spatial scale and temporal resolution at which forest structure and diversity can be assessed. Aerial data appear to offer a promising solution to this problem by enabling rapid capture of forest data at large scales. However, while deep learning methods have been widely used to process these data, current approaches often rely on labels drawn by eye or other subjective reference data, undermining confidence in the reported accuracy of downstream tasks. In this work, I begin to address the validation gap in deep learning-based forest monitoring methods. By developing validation approaches using high-fidelity reference data, and improving methods to automate the generation of such validation data, my thesis establishes a foundation for reliable validation of airborne or satellite-based forest monitoring. I summarise my contributions as follows: I develop a scalable, airborne approach to measuring tree crown dieback and show that when measured end-to-end, the results correlate with ground-based measurements. I use ultra high-fidelity terrestrial LiDAR to derive ground truth annotations for a common intermediate step in such approaches -- aerial instance segmentation -- and use these labels to show that accuracy is falsely inflated when measured using hand-drawn annotations. I develop a new approach based on self-supervised learning to the semantic segmentation of terrestrial LiDAR, and demonstrate that it massively reduces the need for hand-labelling - a major bottleneck to the scaling of contribution 2."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["cdcb0fd52bab0c581eeef2a21af04bb9","87eda9de84448d1f82354d60eee3eb5f"]},{"key":"dc:title","label":"Title","values":["Fidelity versus Scalability: Deep Learning in Forest Remote Sensing"]}]}],"canonical_facts":{"dc:contributor.advisor":["Lines, Emily","Grieve, Stuart"],"dc:contributor.sponsor":["Cambridge Philosophical Society Trinity Hall"],"dc:creator":["Allen, Matthew"],"dc:date.issued":["2025-07-06"],"dc:description.abstract":["Climate change drives complex responses in forest ecosystems, with individual trees responding differently depending on species, size, and local competition. These individual responses to stressors such as drought and temperature can drastically alter forest composition, and cannot be captured through forest-scale assessments. Monitoring forest dynamics at an individual tree scale is therefore essential for accurately assessing ecosystem responses to climate change. Traditional field inventory based on manual measurements is able to provide data at this granularity, but limits the spatial scale and temporal resolution at which forest structure and diversity can be assessed. Aerial data appear to offer a promising solution to this problem by enabling rapid capture of forest data at large scales. However, while deep learning methods have been widely used to process these data, current approaches often rely on labels drawn by eye or other subjective reference data, undermining confidence in the reported accuracy of downstream tasks. In this work, I begin to address the validation gap in deep learning-based forest monitoring methods. By developing validation approaches using high-fidelity reference data, and improving methods to automate the generation of such validation data, my thesis establishes a foundation for reliable validation of airborne or satellite-based forest monitoring. I summarise my contributions as follows: I develop a scalable, airborne approach to measuring tree crown dieback and show that when measured end-to-end, the results correlate with ground-based measurements. I use ultra high-fidelity terrestrial LiDAR to derive ground truth annotations for a common intermediate step in such approaches -- aerial instance segmentation -- and use these labels to show that accuracy is falsely inflated when measured using hand-drawn annotations. I develop a new approach based on self-supervised learning to the semantic segmentation of terrestrial LiDAR, and demonstrate that it massively reduces the need for hand-labelling - a major bottleneck to the scaling of contribution 2."],"dc:format.checksum.md5":["cdcb0fd52bab0c581eeef2a21af04bb9","87eda9de84448d1f82354d60eee3eb5f"],"dc:identifier.doi":["https://doi.org/10.17863/CAM.124542"],"dc:identifier.uri":["https://www.repository.cam.ac.uk/bitstreams/b22c7a5f-6b21-4927-933a-5bb21ca24b15/download"],"dc:language":["eng"],"dc:publisher.institution":["University of Cambridge"],"dc:relation.isreferencedby.uri":["https://www.repository.cam.ac.uk/handle/1810/394756"],"dc:rights":["https://www.repository.cam.ac.uk/bitstreams/398cd768-7262-4888-afd3-4617fdb64a1d/download","http://purl.org/NET/rdflicense/allrightsreserved"],"dc:subject":["Forest Ecology","Climate Change","Deep Learning","Machine Learning","LiDAR","UAV","Airborne","Computer Vision"],"dc:title":["Fidelity versus Scalability: Deep Learning in Forest Remote Sensing"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["Doctoral"],"dc:type.qualificationname":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-07-22T22:24:18Z"}