Sequential Bayesian inference for spatio-temporal models of temperature and humidity data

• Published in: Journal of computational science Vol. 43; p. 101125
• Main Authors: Lai, Yingying, Golightly, Andrew, Boys, Richard J.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.05.2020
• Subjects: Dynamic linear models (DLMs); Iterated batch importance sampling (IBIS); Parallel computing; Sequential Monte Carlo (SMC); Dynamic linear models (DLMs); Parallel computing; Iterated batch importance sampling (IBIS); Sequential Monte Carlo (SMC)
• ISSN: 1877-7503; 1877-7511; 1877-7511
• DOI: 10.1016/j.jocs.2020.101125

•The paper develops a novel spatio-temporal model for temperature and humidity data.•The model is fit to streaming data by using a cutting-edge sequential Monte Carlo algorithm: the iterated batch importance sampling scheme.•We derive an on-line implementation of this algorithm which is more efficient than a standard scheme and also parallelizable.•The on-line algorithm is shown to be more than six times more computationally efficient than a standard implementation.•The model is shown to give a good description of the underlying process and provide reasonable forecast accuracy. We develop a spatio-temporal model to forecast sensor output at five locations in North East England. The signal is described using coupled dynamic linear models, with spatial effects specified by a Gaussian process. Data streams are analysed using a stochastic algorithm, known as iterated batch importance sampling (IBIS), which sequentially propagates a discrete approximation of the parameter posterior through a series of reweighting and resampling steps. To circumvent degeneracy of the parameter samples, additional Markov chain Monte Carlo steps are used, subject to some degeneracy criterion. The IBIS algorithm is modified to make it more efficient and parallisable. The model is shown to give a good description of the underlying process and provide reasonable forecast accuracy.