White Rose University Consortium logo
University of Leeds logo University of Sheffield logo York University logo

Bayesian Inference for Dynamic Spatio-temporal Models

Karadimitriou, Sofia Maria (2018) Bayesian Inference for Dynamic Spatio-temporal Models. PhD thesis, University of Sheffield.

[img] Text
Thesis final.pdf
Restricted until 21 September 2020.

Request a copy


Spatio-temporal processes are phenomena evolving in space, either by being a point, a field or a map and also they vary in time. A stochastic process may be proposed as a vehicle to infer and hence offer predictions of the future. In this era high dimensional datasets can be available where measurements are observed daily or even hourly at more than one locations along with many predictors. Therefore, what we would like to infer is high dimensional and the analysis is difficult to come through due to high complexity of calculations or efficiency from a computational aspect. The first Reduced-dimension Dynamic Spatio Temporal Models (DSTMs) were developed to jointly describe the spatial and temporal evolution of a function observed subject to noise. A basic state space model is adopted for the discrete temporal variation, while a continuous autoregressive structure describes the continuous spatial evolution. Application of DTSMs rely upon the pre-selection of a suitable reduced set of basis functions and this can present a challenge in practice. In this thesis we propose a Hierarchical Bayesian framework for high dimensional spatio-temporal data based upon DTSMs which attempts to resolve this issue allowing the basis to adapt to the observed data. Specifically, we present a wavelet decomposition for the spatial evolution but where one would typically expect parsimony. This believed parsimony can be achieved by placing a Spike and Slab prior distribution on the wavelet coefficients. The aim of using the Spike and Slab prior, is to filter wavelet coefficients with low contribution, and thus achieve the dimension reduction with significant computational savings. We then propose an Hierarchical Bayesian State-space model, for the estimation of which we offer an appropriate Forward Filtering Backward Sampling algorithm under an MCMC procedure. Then, we extend this model for estimating Poisson counts and Multinomial cell probabilities through proposing a Conditional Particle Filtering framework.

Item Type: Thesis (PhD)
Academic Units: The University of Sheffield > Faculty of Science (Sheffield) > School of Mathematics and Statistics (Sheffield)
Depositing User: Miss Sofia Maria Karadimitriou
Date Deposited: 25 Mar 2019 10:06
Last Modified: 25 Mar 2019 10:06
URI: http://etheses.whiterose.ac.uk/id/eprint/23358

Please use the 'Request a copy' link(s) above to request this thesis. This will be sent directly to someone who may authorise access.
You can contact us about this thesis. If you need to make a general enquiry, please see the Contact us page.

Actions (repository staff only: login required)