Creating an LSST deblender for low-surface-brightness science

Our in-kind contribution (UKD-UKD-S6) will produce low surface brightness (LSB) friendly images. However, for these images to be scientifically useful, the flux preservation (which is the aim of our work) has to be followed by successful deblending and object detection (which was not a proposed part of our contribution, since the focus is on sky subtraction which is a complex task that absorbs all our bandwidth).

The default LSST deblender (Scarlet) has been designed to work on (the currently default) LSST images which are the result of aggressive sky subtraction which removes LSB flux. Tests show that running this version of Scarlet on LSB friendly images currently fails because LSB flux connects the vast majority of objects, making it difficult to cleanly separate sources in the images. However, deblending and object detection are critical for producing object catalogues which will underpin science. Thus, at it stands now, there will be a way to create LSST images that preserves LSB flux (through our work) but no way to create object catalogues on which the scientific exploitation of these images hinges.

In that context, Scarlet will essentially have to be adapted to work with LSB friendly images.

The named PDRA (Dr Thomas Sedgwick) will work intensively alongside Fred Moolekamp (LSST Data Management) to perform the following steps:
• Test Scarlet on LSB-friendly images (e.g. from HSC) to identify the issues.
• Design metrics that quantify the performance of Scarlet on LSB friendly images.
• Identify potential modifications that would enable Scarlet to operate effectively on LSB friendly images from LSST.