Revisiting the Emission Line Source Detection Problem in Integral Field Spectroscopic Data

TL;DR

This paper introduces a 3D matched filtering method for detecting faint emission-line sources in integral-field spectroscopic data, improving detection significance and enabling analytical derivation of the selection function, validated with real survey data.

Contribution

The authors develop a new 3D matched filtering technique tailored for variable background noise in spectroscopic data, with an analytical approach to determine the selection function, validated on real datasets.

Findings

Enhanced detection significance for faint sources near atmospheric OH bands

Analytical derivation of the selection function from data variances

Validated method with real survey data and source recovery experiments

Abstract

We present a 3-dimensional matched filtering approach for the blind search of faint emission-line sources in integral-field spectroscopic datasets. The filter is designed to account for the spectrally rapidly varying background noise due to the telluric air glow spectrum. A software implementation of this matched filtering search is implemented in an updated version of the Line Source Detection Cataloguing tool (LSDCat2.0). Using public data from the MUSE-Wide survey we show how the new filter design provides higher detection significances for faint emission line sources buried in between atmospheric [OH]-bands at $\lambda \gtrsim 7000$\,\AA{}. We also show how, for a given source parameterisation, the selection function of the improved algorithm can be derived analytically from the variances of the data. We verify this analytic solution against source insertion and recovery experiments in the recently released dataset of the MUSE eXtreme Deep Field (MXDF). We then illustrate how the selection function has to be re-scaled for 3D emission line source profiles that are not fully congruent with the template. This procedure alleviates the construction of realistic selection functions by removing the need for computationally cumbersome source insertion and recovery experiments.