Dynamic normalization for compact binary coalescence searches in non-stationary noise

The output of gravitational-wave interferometers, such as LIGO and Virgo, can be highly non-stationary. Broadband detector noise can affect the detector sensitivity on the order of tens of seconds. Gravitational-wave transient searches, such as those for colliding black holes, estimate this noise in order to identify gravitational-wave events. During times of non-stationarity we see a higher rate of false events being reported. To accurately separate signal from noise, it is imperative to incorporate the changing detector state into gravitational-wave searches. We develop a new statistic which estimates the variation of the interferometric detector noise. We use this statistic to re-rank candidate events identified during LIGO-Virgo's second observing run by the PyCBC search pipeline. This results in a 5% improvement in the sensitivity volume for low mass binaries, particularly binary neutron stars mergers.