A comparison of methods to characterize non-stationarity in time series data

In this work we examine methods to characterize a time-dependent noise background of time series data generated by the Laser Interferometer Gravitational wave Observatories (LIGO). This non-stationarity originates from both instrumental and environmental noise, and can be exhibited by sharp transient features in the data as well as by slowly-varying statistical properties. The efficient identification of the presence of non-stationarity can have the net effect of increasing the sensitivity of the detectors. We present the results of various methods we applied to characterize the non-stationarity of the data, including machine-learning approaches. These methods have exhibited significant overlap with results generated by standard LIGO data monitoring tools.