Probabilistic methods of analysis for the time series Moran scatterplot quadrant signature

Rohde, D., Corcoran, J., McGee, T. R., Wickes, R. and Townsley, M. (2014) Probabilistic methods of analysis for the time series Moran scatterplot quadrant signature. Environmetrics, 26 1: 52-65. doi:10.1002/env.2302

Author Rohde, D.
Corcoran, J.
McGee, T. R.
Wickes, R.
Townsley, M.
Title Probabilistic methods of analysis for the time series Moran scatterplot quadrant signature
Journal name Environmetrics   Check publisher's open access policy
ISSN 1180-4009
Publication date 2014-09-08
Year available 2014
Sub-type Article (original research)
DOI 10.1002/env.2302
Open Access Status
Volume 26
Issue 1
Start page 52
End page 65
Total pages 14
Place of publication West Sussex, United Kingdom
Publisher John Wiley and Sons Ltd
Collection year 2015
Language eng
Formatted abstract
Recently, time series Moran scatterplot quadrant signatures (MSQS) have been introduced as a means of spatio-temporal analysis. Moran scatter plots summarise a set of local Moran statistics that identify how a quantity of interest relates to its neighbours. By reducing these scatter plots to just one of four quadrant locations at fixed time intervals, a time series MSQS is obtained. Clustering on the time series MSQS allows regions that show similar behaviour relative to their neighbours to be identified. It has recently been shown that the Levenshtein metric, a distance metric originally used for string comparison, can be used to construct a kernel that allows standard Ward hierarchical clustering to be applied. The result is that regions with similar spatio-temporal behaviour can be identified. The purpose of this paper is to demonstrate the use of Dirichlet Process mixture models as a fully probabilistic alternative to applying Ward hierarchical clustering with the Levenshtein metric. This approach offers an advance to the existing literature as it proposes a fully generative model that articulates the underlying assumptions and allows prediction of new test points. Additionally, it provides a principled method that avoids using heuristics to select the number of clusters. An efficient Gibbs sampling Markov chain Monte Carlo algorithm is presented, and it is demonstrated how the output of this can be mapped and analysed. Difficulties inherent in mapping and plotting high-dimensional mathematical objects are discussed, and practical solutions are proposed.
Keyword Bayesian statistics
Machine learning
Spatial statistics
Fire modelling
Dirichlet process
Q-Index Code C1
Q-Index Status Confirmed Code
Institutional Status UQ

Version Filter Type
Citation counts: TR Web of Science Citation Count  Cited 0 times in Thomson Reuters Web of Science Article
Scopus Citation Count Cited 0 times in Scopus Article
Google Scholar Search Google Scholar
Created: Wed, 26 Nov 2014, 09:48:46 EST by Helen Smith on behalf of School of Geography, Planning & Env Management