Impact of noise transients on low latency gravitational-wave event localization

Gravitational-wave (GW) data contains non-Gaussian noise transients called “glitches.” During the third LIGO-Virgo observing run about 24% of all gravitational-wave candidates were in the vicinity of a glitch, while even more events could be affected in future observing runs due to increasing detector sensitivity. This poses a problem since glitches can affect the estimation of GW source parameters, including sky localization, which is crucial to identify an electromagnetic counterpart. This is the first of its kind study that evaluates the importance of relative glitch positioning in time with respect to a GW signal. In this paper we estimate how much sky localization is affected by a nearby glitch in low latency. We injected binary black hole (BBH), binary neutron star (BNS), and neutron star-black hole (NSBH) signals into data containing three different classes of glitches: blips, thunderstorms and fast scatterings. The impact of these glitches was assessed by estimating the number of tile pointings that a telescope would need to search over until the true sky location of an event is observed. We find that blip glitches generally do not affect the localization of our tested GW signals, however in very rare cases of a blip glitch overlap with a BBH or a NSBH signal can cause the true position of the event to lie well outside the 90% computed sky localization, severely compromising electromagnetic follow-up. Thunderstorm glitches have a noticeable impact on BBH and NSBH events, especially if there is no third interferometer. In such cases we find that the electromagnetic follow-up efforts with telescopes as large as $20{\mathrm{deg}}^2$ field of view (FOV) are affected. Observing BBH and NBSH signals with three-detector network reduces the bias in sky localization caused by thunderstorm glitches, making the bias to affect only small ($\mathrm{FOV}=1{\mathrm{deg}}^2$) telescopes. BNS events appear to be not affected by thunderstorm glitches. Fast scattering glitches have no impact on the low latency localization of BBH and BNS signals. For NSBH signals observed with two-detector network, the sky localization bias due to fast scattering glitches is significant enough to affect even large ($\mathrm{FOV}=20{\mathrm{deg}}^2$) telescopes. Observing NSBH signals with three interferometers reduces the bias such that it impacts only small ($\mathrm{FOV}=1{\mathrm{deg}}^2$) telescopes.

Figure 1

Time-frequency representations of three classes of glitches considered in our study. Blip glitches (a) are usually subsecond in duration and have a wide frequency bandwidth, $\mathcal{O}(100)\mathrm{Hz}$. Thunderstorm glitches (b) are multiple seconds ($3–10\mathrm{s}$) in length and have a frequency of less than 200 Hz. Fast scattering glitches (c) are often found in groups and can last up to minutes with a typical frequency of 20–70 Hz. Note the different time and frequency axis values for each figure.

Figure 2

Average number of tiles required to observe the true sky location of a GW150914-like ${\mathrm{SNR}}_{\mathrm{Net}}=27$ event relative to the blip glitch central time $t_0$. For both FOV telescopes, small ($1{\mathrm{deg}}^2$, orange) and large ($20{\mathrm{deg}}^2$, blue), tiling deficiency at $t_0+30\mathrm{ms}$ is observed. Shaded bands represent $1\sigma$ deviation. Note that these are averaged results from 10 blip glitches with 20% trimming.

Figure 3

Time series data of a GW150914-like injected signal (blue) in Livingston data (orange) containing a blip glitch. The GW signal is injected at $t_0+30\mathrm{ms}$ relative to the blip glitch central time $t_0$. The glitch is aligned with the peak GW signal in a such way that they have opposite phases.

Figure 4

Sky localization of a GW150914-like event injected at $t_0+30\mathrm{ms}$ relative to the blip glitch central time $t_0$. The skymap of the time series data from Fig. 3 (top) shows that the true sky location of the event (indicated by blue pointer) does not coincide with the estimated sky location. This happens because the glitch overlaps and cancels part of the GW signal. If the same GW signal is phase-shifted by $\pi /2$, the true event sky location agrees well with the estimated sky location (bottom).

Figure 5

Average number of tiles required to observe the true sky location of a GW190814-like event for ${\mathrm{SNR}}_{\mathrm{Net}}=27$. For both FOV telescopes, small ($1{\mathrm{deg}}^2$, orange) and large ($20{\mathrm{deg}}^2$, blue), the tiling deficiency at $t_0+25\mathrm{ms}$ is observed. Shaded bands represent $1\sigma$ deviation. Note that these are averaged results from 10 blip glitches with 20% trimming.

Figure 6

Average number of tiles required to observe the true sky location of a GW170817-like ${\mathrm{SNR}}_{\mathrm{Net}}=27$ event relative to the blip glitch central time $t_0$. There is no noticeable change in the tiling efficiency for both FOV telescopes, small ($1{\mathrm{deg}}^2$, orange) and large ($20{\mathrm{deg}}^2$, blue). Shaded bands represent $1\sigma$ deviation. Note that these are averaged results from 10 blip glitches with 20% trimming.

Figure 7

Time-frequency representation of a GW170817-like signal overlapping a thunderstorm glitch. Only a minor part (about 1.7 s in total) of the BNS signal coincides with a thunderstorm glitch.