Identification of regime shifts in time series using neighborhood statistics

Sudden and significant changes in biotic and abiotic variables have been observed across a variety of systems. The identification of such regime shifts in time series includes both model-fitting and statistical approaches. We introduce two methods that use state- or measurement-space neighborhood statistics to pick out regime shifts. Analysis of simulated and real data sets shows that these methods can be an effective means of identifying regime shifts for single variable as well as multivariable time series. In addition, these methods can be used on systems with non-equilibrium steady states. However, care must be taken in interpreting results as these methods do respond to changes in time series that are not consistent with the regime shift concept.