When is Stacking Confusing?: The Impact of Confusion on Stacking in Deep HI Galaxy Surveys

TL;DR

This paper develops an analytic model to predict how confusion from overlapping sources affects HI stacking in galaxy surveys, highlighting the importance of resolution and survey parameters.

Contribution

It introduces a new analytic model to quantify confusion effects in HI stacking surveys, validated against existing data and used to forecast limitations for future surveys.

Findings

Confusion impacts are minimal for high-resolution surveys like LADUMA and DINGO UDEEP.

Lower resolution surveys like FAST will face significant confusion issues beyond z=0.1.

Confusion signals can form a double Gaussian profile, affecting interpretation of stacked data.

Abstract

We present an analytic model to predict the HI mass contributed by confused sources to a stacked spectrum in a generic HI survey. Based on the ALFALFA correlation function, this model is in agreement with the estimates of confusion present in stacked Parkes telescope data, and was used to predict how confusion will limit stacking in the deepest SKA-precursor HI surveys. Stacking with LADUMA and DINGO UDEEP data will only be mildly impacted by confusion if their target synthesised beam size of 10 arcsec can be achieved. Any beam size significantly above this will result in stacks that contain a mass in confused sources that is comparable to (or greater than) that which is detectable via stacking, at all redshifts. CHILES' 5 arcsec resolution is more than adequate to prevent confusion influencing stacking of its data, throughout its bandpass range. FAST will be the most impeded by confusion, with HI surveys likely becoming heavily confused much beyond z = 0.1. The largest uncertainties in our model are the redshift evolution of the HI density of the Universe and the HI correlation function. However, we argue that the two idealised cases we adopt should bracket the true evolution, and the qualitative conclusions are unchanged regardless of the model choice. The profile shape of the signal due to confusion (in the absence of any detection) was also modelled, revealing that it can take the form of a double Gaussian with a narrow and wide component.