Change Detection in Time Series

Time series are analysed to detect changes and predict future behaviour. Monitoring of the surface or subsurface in Geoscience provides such time series. As new data becomes available, changes of regime are detected. They should be identified as early as possible with as few false positive as possible. In this paper, we define the ideas of regime and change of regime in a time series. We then give an overview of the Bayesian method used to detect these changes. The principles are illustrated with an application for the detection of a mine tailings dam failure using InSAR satellite data.