Upton, Maeve (2023) Bayesian generalised additive models for quantifying sea-level change: Methods and Software. PhD thesis, National University of Ireland Maynooth.

Download (31MB)

Abstract

Rising sea levels pose significant risks to coastal regions worldwide, and the 2021 Intergovernmental Panel on Climate Change AR6 report emphasised that rates of sea-level rise are the fastest in at least the last 3000 years. To understand historical sea-level trends at regional and local scales, it is crucial to analyse the drivers of sea-level change and their potential impacts. The influence of these different drivers interact at a range of spatial (global, regional, local level) and temporal (annual to millennia) scales. The development of a statistical model that seeks to estimate a number of these characteristics would be of immeasurable value to the sea level and climate impact communities. These characteristics would include: exhibiting flexibility in time and space; having the capability to examine the separate drivers; and taking account of uncertainty. The aim of our project is to develop statistical models to examine historic sea-level changes for North America’s Atlantic coast and extend to the North Atlantic region, incorporating Ireland’s coastline. For our models, we utilise sea-level proxies and tide gauge data which provide relative sea level estimates with uncertainty. Proxy data can reconstruct sea-level variations over the late Holocene, spanning the last 2000 years, providing a valuable pre-anthropogenic context for understanding historical relative sea-level changes. We study a range of statistical models used to examine relative sea-level data accounting for uncertainty and varying in space and time. The statistical approaches employed range from simple linear regressions to advanced Bayesian Generalised Additive Models (GAMs), which allow separate components of sea-level change to be modelled individually and efficiently and for smooth rates of change to be calculated. Our most advanced models are built in a Bayesian framework which allows for external prior information to constrain the evolution of sea-level change over space and time. To investigate the drivers of sea-level change, we use flexible and extended GAMs and effectively account for the uncertainty associated with proxy data using the noisy input uncertainty method. Through the integration of statistical models, proxy data, and tide gauge measurements, our findings reveal a significant rise in current sea levels along North America’s Atlantic coast, reaching the highest point in at least the last 15 centuries. The GAMs exhibit a remarkable capability to examine various drivers of relative sea level change, including geological processes (e.g. glacial isostatic adjustment; GIA), local factors, and barystatic influences. Our models provide evidence that GIA primarily drove relative sealevel change along North America’s Atlantic coast until the 20th century when a notable rise in the rate of sea-level rise became apparent. We present the open-source reslr package, which serves as a valuable resource for the sea level community, offering a diverse range of statistical approaches. This R package enables Bayesian modeling of relative sea level data, providing a unified framework for loading data, fitting models, and summarising results. By incorporating various statistical models, it offers flexibility and versatility in sea level analysis. Notably, reslr takes into account measurement errors associated with relative sea-level data in multiple dimensions, enhancing the accuracy and reliability of the modelling process. With reslr, researchers and practitioners can explore and compare different statistical methodologies for a comprehensive understanding of historical sea-level changes, their uncertainties and importantly, the rate of change of these sea-level variations. One critical driver of sea-level change is ocean dynamics, commonly referred to as dynamic sea-level change. Our statistical methodologies offer valuable insights into dynamic sea-level changes over the last 2,000 years, using both proxy records and tide gauges at a regional level. To investigate the dynamic sea-level component along the North Atlantic coastline, we employ an extended noisy input GAM, effectively decomposing the relative sea-level signal. In our investigation, we focus on two key components of dynamic sea-level change in the North Atlantic: the vertical (Atlantic Meridional Overturning Circulation - AMOC) circulation and the quasi-horizontal circulation, involving surface-enhanced currents and gyres. Our results highlight a decline in the AMOC over the studied period of 2,000 years with an unprecedented rate of decrease similar to previous studies. Additionally, the quasi-horizontal circulation exhibits increased variability during the same timeframe with a notably difference north and south of Cape Hatteras, USA. This comprehensive analysis sheds light on the complex dynamics driving sea-level changes in the North Atlantic region, contributing to a better understanding of the factors influencing sea-level variations. Our approach places the present alterations in ocean circulation patterns within the extended context of a 2,000-year timeframe. However, the interpretability of these changes is constrained by the resolution of the proxy data.

Item Type:	Thesis (PhD)
Keywords:	Bayesian generalised additive models; quantifying sea-level change; Methods and Software;
Academic Unit:	Faculty of Science and Engineering > Research Institutes > Hamilton Institute Faculty of Social Sciences > Research Institutes > Irish Climate Analysis and Research Units, ICARUS
Item ID:	18319
Depositing User:	IR eTheses
Date Deposited:	26 Mar 2024 15:13
URI:
Use Licence:	This item is available under a Creative Commons Attribution Non Commercial Share Alike Licence (CC BY-NC-SA). Details of this licence are available here

Repository Staff Only(login required)

Item control page

Downloads

Origin of downloads