Data-driven approaches to coastal resilience using monitoring and machine learning

Murphy, James (2025) Data-driven approaches to coastal resilience using monitoring and machine learning PhD thesis, University of Liverpool.

Text 200990852_Mar2025.pdf - Author Accepted Manuscript Download (3MB) | Preview

Abstract

Coastal environments face growing threats from human activity that disrupts natural processes and hazards related to climate change, such as rising sea levels. Evaluating the resilience of coastal cities - their capacity to endure and bounce back from these challenges - has become essential. A key part of this assessment is the capacity to observe coastal environments and understand their dynamics and how they change over time. Recent advancements in technology and data analysis methods enable the use of high-resolution datasets to investigate the interactions between coastlines and coastal system populations. While existing research on coastal resilience has established a foundation for data-driven policy and decision-making, it often lacks in-depth studies that integrate physical and social factors. This thesis addresses this gap by utilising innovative coastal monitoring approaches that combine data analysis with machine-learning techniques to evaluate coastal processes and their interactions with human activity. Additionally, it emphasises the importance of studying short-term beach dynamics to enhance our understanding of their cumulative long-term impacts. This thesis presents a series of scalable methodologies and analytical toolkits for comprehensive spatiotemporal analysis of human and physical processes along the coast, utilising data from the northwest coast of England and the eastern coast of Australia. By utilising marine radar data, open-source datasets, and custom machine learning models, it makes three main contributions: (1) it characterises beach morphology through bi-weekly digital elevation models generated from radar data, which capture short-term changes and their impact on long-term processes; (2) it monitors human movement across beach areas by applying machine learning models to data obtained with a marine radar system, offering insights into beach user behaviour that aid safety, recreation, and conservation; and (3) it integrates 30 years of population density and shoreline data with high-resolution temporal clustering algorithms to evaluate the interactions between coastal populations and physical processes, providing crucial insights for adaptable policies aimed at building resilient coastal cities. Firstly, the spatiotemporal analysis of elevation change using radar data at Rossall Beach, UK, revealed a wide range of heterogeneous changes that followed a pattern where greater fluctuations in beach elevation occurred at the tidal extremes, evidenced by a significantly higher standard deviation at the lower and upper tide lines, particularly around the beach groynes. The developed tools also indicated that these changes are influenced by storm events, showing that substantial erosive periods are often followed by rapid recovery in subsequent surveys. Secondly, by repurposing the same raw radar outputs that generate the derived DEMs, human movement could be tracked across the beach environment at Crosby Beach, UK, through the training of machine learning classifiers. In this instance, a Support Vector Classifier (SVC), a Logistic Regression (LR) model, and a Random Forest (RF) model were trained on human trajectory data, and their outputs were compared. The random forest classifier yielded accurate and favourable results, successfully identifying numerous human trajectories while maintaining a low rate of false positives. Subsequently, this model was tested on unseen data and the density of human movement at Crosby Beach over a two-hour period was shown as a proof of concept. Lastly, by integrating and clustering population density and shoreline data in the City of Gold Coast, Australia, a representation of coastal hazard exposure between 1990 and 2020 was constructed on a 100m x100m grid. Consequently, it is possible to observe how human populations, and the physical dynamics of the coastline interact within a rapidly expanding coastal city. Specifically, areas such as Surfers Paradise are at an increasing risk of coastal hazard exposure, with high-density populations in regions that have exhibited ongoing erosion, leading to higher vulnerability. Conversely, it is evident how relative coastal stability has facilitated population growth in other areas, such as Southport and Paradise Point. Key contributions of this research arise from a) the use of unique datasets derived from marine radar and b) the analysis of intricate spatiotemporal trends, utilising data with a high temporal and spatial resolution. Moreover, the application of tailored machine learning models to extract meaningful patterns and trends from these extensive spatial datasets opens up new opportunities for coastal monitoring, providing significant value to coastal managers and stakeholders, such as government agencies, environmental groups, and academic institutions.

Item Type:	Thesis (PhD)
Uncontrolled Keywords:	coastal resilience, data analysis, machine learning, radar, remote sensing
Divisions:	Faculty of Science & Engineering Faculty of Science & Engineering > School of Environmental Sciences
Depositing User:	Symplectic Admin
Date Deposited:	15 Aug 2025 09:36
Last Modified:	01 Feb 2026 02:30
DOI:	10.17638/03192756
Supervisors:	Arribas-Bel, Daniel Plater, Andy
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3192756
Disclaimer:	The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.