Bryce, Erin (2025) An exploration of spatio-temporal statistical models for landslide hazard assessment and prediction. PhD thesis, University of Glasgow.
Full text available as:![]() |
PDF
Download (97MB) |
Abstract
This thesis develops and applies statistical modelling techniques for complex spatial and spatio-temporal geophysical datasets, with a particular focus on the estimation and assessment of landslide hazard and surface deformation. The bulk of this thesis’s methodological framework is grounded in Bayesian inference, utilising the integrated nested Laplace approximation (INLA) for efficient computation. The models employed are of the latent Gaussian class, wherein observations are conditioned on an unobserved latent field that captures residual spatial or spatio-temporal variation. This latent field is represented via a Mat´ern Gaussian Random Field, approximated through the Stochastic Partial Differential Equation (SPDE) approach. Domain-specific covariates - geographical, geological and meteorological - are incorporated within the hierarchical structure. In doing so, the thesis explores landslide hazard in terms of where landslides occur, when they occur, and how large they are.
The statistical approaches developed span a range of modelling strategies, including susceptibility models (presence/absence), Poisson and log-Gaussian Cox processes (LGCPs), functional generalised additive models (FGAMs), and a custom space-time SPDE smoother implemented within the Mixed GAM Computation Vehicle with Automatic Smoothness Estimation package (mgcv), which is a flexible framework for modelling non-linear relationships within GAMs. This enables the integration of high-resolution environmental covariates and a functional precipitation predictor, with various continuous distributions used to model landslide size.
Chapter 2 introduces a unified landslide hazard framework through a Hurdle model, jointly modelling occurrence (via a Bernoulli process) and size (via a log-Gaussian model for planimetric extent). This enables the creation of hazard maps that provide probabilistic estimates of large-event exceedances, along with their associated uncertainty.
Chapter 3 presents an updated landslide susceptibility model for Scotland, developed for the British Geological Survey (BGS). It includes a proposed LGCP extension and provides the first data-driven landslide susceptibility framework for the BGS, benchmarked against their previous heuristic model, GeoSure.
Chapter 4 addresses a key methodological challenge: the influence of mesh resolution and integration scheme in SPDE-based point process models with fine-scale covariates. Motivated by issues encountered in the BGS application, a series of simulation studies explores the effects of mesh specification, culminating in a case study where a marked LGCP is fitted to a Japanese landslide inventory using landslide size as the mark.
Chapter 5 explores the temporal dimension of landslide hazard through a spatio-temporal model of surface deformation in a region of China over a two-month period. This chapter introduces a functional precipitation predictor and transitions from a Bayesian to a frequentist framework, motivated by limitations in earlier precipitation representations. It implements, for the first time, a space-time SPDE Matern smoother within mgcv, enabling flexible modelling of deformation using high-resolution covariates and functional data analysis techniques.
Item Type: | Thesis (PhD) |
---|---|
Qualification Level: | Doctoral |
Additional Information: | Supported by funding from the Additional Funding Programme for Mathematical Sciences, delivered by EPSRC (EP/V521917/1) and the Heilbronn Inst. |
Subjects: | G Geography. Anthropology. Recreation > GE Environmental Sciences H Social Sciences > HA Statistics Q Science > QA Mathematics |
Colleges/Schools: | College of Science and Engineering > School of Mathematics and Statistics |
Funder's Name: | Engineering and Physical Sciences Research Council (EPSRC), EP/V521917/1 |
Supervisor's Name: | Castro-Camilo, Dr. Daniela, Illian, Dr. Janine and Lombardo, Dr. Luigi |
Date of Award: | 2025 |
Depositing User: | Theses Team |
Unique ID: | glathesis:2025-85477 |
Copyright: | Copyright of this thesis is held by the author. |
Date Deposited: | 26 Sep 2025 10:07 |
Last Modified: | 26 Sep 2025 10:11 |
Thesis DOI: | 10.5525/gla.thesis.85477 |
URI: | https://theses.gla.ac.uk/id/eprint/85477 |
Related URLs: |
Actions (login required)
![]() |
View Item |
Downloads
Downloads per month over past year