UNIONS and Euclid: Calibrating point source--galaxy separation for ground-based surveys
Abstract
An efficient and accurate separation of unresolved point sources from extended objects, usually referred to as star–galaxy separation, is essential for the scientific exploitation of deep wide-field imaging surveys, particularly at faint magnitudes where morphological differences become subtle. This paper presents a detailed assessment of two morphology-based classification methods applied to single-band, ground-based imaging from the Ultraviolet Near-Infrared Optical Northern Survey (UNIONS), using observations as a high-fidelity reference to assess their performance, and to train one of the methods. Euclid We analysed 8 deg2$ of overlapping UNIONS r-band and Euclid data, comprising more than 7 data enable robust morphological labelling down to sources. The Euclid 26, which we used to train and validate two classification approaches applied to UNIONS alone; i.e. a curve-of-growth method and a supervised neural-network classifier, both of which are based on aperture photometry and image-quality information. The curve-of-growth method yields an almost perfectly complete point-source sample ($>95%$), but its purity decreases rapidly at magnitudes fainter than r ≃ 22.5. The neural-network approach significantly improves the purity at the faint end, maintaining ≥ 80% purity down to r ≃ 24, while the completeness decreases to almost $50% at similar depths. In both cases, the classification performance strongly depends on the image quality of the ground-based data. This work demonstrates how data can be used to develop, calibrate, and validate star–galaxy separation methods for ground-based surveys, which are particularly useful in regions not covered by . Both methods have already been applied to UNIONS data and are readily applicable to future surveys such as LSST, providing flexible solutions depending on whether completeness or purity is prioritised. Euclid Euclid