Parameter estimation with non stationary noise in gravitational waves data

The sensitivity of gravitational-waves detectors is characterized by their noise curves which determine the detector's reach and the ability to accurately measure the parameters of astrophysical sources. The detector noise is typically modelled as stationary and Gaussian for many practical purposes. However, physical changes in the state of detectors due to environmental and instrumental factors, including extreme cases where a detector discontinues observing for some time, introduce non-stationarity into the noise. Even slow evolution of the detector sensitivity will affect long duration signals such as binary neutron star (BNS) mergers. Mis-estimation of the noise behavior directly impacts the posterior width of the signal parameters. This becomes an issue for studies which depend on accurate localization volumes such as i) probing cosmological parameters (such as Hubble constant, clustering bias) using cross-correlation methods with galaxies, ii) doing electromagnetic follow-up using localization information from parameter estimation done from pre-merger data. We study the effects of dynamical noise on the parameter estimation of the GW events. We develop a new method to correct dynamical noise by estimating a locally-valid pseudo PSD which is normalized along the time-frequency track of a potential signal. We do simulations by injecting the BNS signal in various scenarios where the detector goes through a period of non-stationarity with reference noise curve of third generation detectors (Cosmic explorer, Einstein telescope). As an example, for a source where mis-modelling of the noise biases the signal-to-noise estimate by even $10\%$, one would expect the estimated localization volume to be either under or over reported by $\sim 30\%$; errors like this, especially in low-latency, could potentially cause follow-up campaigns to miss the true source location.