Sky subtraction techniques: new paper published

MNRAS has published a paper by LSST: UK team member Dr Aaron Watkins and collaborators which examines the merits and pitfalls of different kinds of sky subtraction techniques. Strategies for optimal sky subtraction in the low surface brightness regime describes experiments using fully synthetic images to investigate three different techniques, two commonly used in low surface brightness surveys, another experimental.  

The study's results quantify the impact of undetected flux on sky models, which tends to bias estimated sky brightnesses high, risking over-subtraction of flux. If the sky is modelled with a complex function (for example, a high-order polynomial, or a spline interpolation), that over-subtraction can occur locally, leading to artificial divots surrounding extended objects like galaxies, or even objects which are simply located close together on the sky. However, the results demonstrate that when a simple model is used, and when proper care is taken to mask detected astronomical objects to low surface brightness levels, this bias can be reduced to negligible amounts.  

Even for a survey as deep as LSST, any sky subtraction technique traditionally used in low surface brightness surveys can still be applied safely, so long as empirical corrections are made for scattered light and undetected faint sources.

The paper justifies the recommendations Watkins and collaborators have proposed to LSST's data management (DM) team regarding the survey's pipeline sky subtraction. Working alongside DM, Watkins et al. found that the existing algorithm suffers from two problems: insufficient masking of low surface brightness flux, and too complex a sky model.  Following the paper's results, the team found that adjusting the algorithm to use a much simpler model, even without an improvement to the masking, proves very successful at minimising the impact of the sky subtraction on the flux of extended or clustered objects.  

In fact, a preliminary investigation suggests that the proposed revised algorithm might benefit more than just low surface brightness science: a number of DM's photometric quality metrics appear to improve slightly when the revised algorithm is used, compared to the default pipeline.  However, the full impact of the proposed change is still being investigated.