Hierarchical Spatio-Temporal Change-Point Detection

Abstract

Detecting change-points in multivariate settings is usually carried out by analyzing all marginals either independently, via univariate methods, or jointly, through multivariate approaches. The former discards any inherent dependencies between different marginals and the latter may suffer from domination/masking among different change-points of distinct marginals. As a remedy, we propose an approach which groups marginals with similar temporal behaviors, and then performs group-wise multivariate change-point detection. Our approach groups marginals based on hierarchical clustering using distances which adjust for inherent dependencies. Through a simulation study we show that our approach, by preventing domination/masking, significantly enhances the general performance of the employed multivariate change-point detection method. Finally, we apply our approach to two datasets: (i) Land Surface Temperature in Spain, during the years 2000–2021, and (ii) The WikiLeaks Afghan War Diary data.

KEYWORDS:

• Clustering
• Functional data
• Land surface temperature
• Multivariate analysis
• Point patterns
• Satellite images
• Trace-variogram

Supplementary Materials

The supplementary material contains additional background to the manuscript, including some preliminaries on multivariate change-point detection, and hierarchical clustering with details regarding linkage functions, and methods to estimate optimal clusters for spatially dependent data in Section S1. Additional simulation studies can be found in Section S2 which cover various scenarios concerning changes in the mean, the variance, multiple changes, and scalability of HSTCPD. Further descriptions and results about our real datasets (i.e. LST in Spain from February 2000 to November 2021, and AWD data which runs monthly from January 2004 to December 2009) are also presented in Section S3.

Acknowledgments

The authors are grateful to the editor and two reviewers for useful comments. The authors are also grateful to Yufeng Liu for supplying the R code of the AdaptiveCpt method. All the presented results are reproducible, and the corresponding R code can be found at https://github.com/Moradii/HSTCPD.

Disclosure Statement

The authors report there are no competing interests to declare.