Deconstructing a galaxy: colour distributions of point sources in Messier 83

TL;DR

This study analyzes the colour distributions of nearly 70,000 point sources in Messier 83 using multiband photometry, exploring clustering methods and their effectiveness in classifying different stellar objects.

Contribution

It evaluates the effectiveness of clustering algorithms on multiband colour data to classify sources in a nearby galaxy, providing insights for future galaxy surveys.

Findings

K-Means identified likely star clusters.

Mean Shift detected outliers at colour distribution edges.

Combining narrow-band data offers limited benefit for spectral line source identification.

Abstract

What do we see when we look at a nearby, well-resolved galaxy? Thousands of individual sources are detected in multiband imaging observations of even a fraction of a nearby galaxy, and characterizing those sources is a complex process. This work analyses a ten-band photometric catalogue of nearly 70 000 point sources in a 7.3 square arcmin region of the nearby spiral galaxy Messier 83, made as part of the Early Release Science programme with the Hubble Space Telescope's Wide Field Camera 3. Colour distributions were measured for both broad-band and broad-and-narrow-band colours; colours made from broad-bands with large wavelength differences generally had broader distributions although B - V was an exception. Two- and three-dimensional colour spaces were generated using various combinations of four bands and clustered with the K-Means and Mean Shift algorithms. Neither algorithm was able to consistently segment the colour distributions: while some distinct features in colour space were apparent in visual examinations, these features were not compact or isolated enough to be recognized as clusters in colour space. K-Means clustering of the UBVI colour space was able to identify a group of objects more likely to be star clusters. Mean Shift was successful in identifying outlying groups at the edges of colour distributions. For identifying objects whose emission is dominated by spectral lines, there was no clear benefit from combining narrow-band photometry in multiple bands compared to a simple continuum subtraction. The clustering analysis results are used to inform recommendations for future surveys of nearby galaxies.