First Look at Photometric Reduction via Mixed-Model Regression (Poster abstract)

Volume 44 number 2 (2016)

Eric Dose
4021 SW 10th Avenue #412, Topeka, KS 66604; astro@ericdose.com

Abstract

(Abstract only) Mixed-model regression is proposed as a new approach to photometric reduction, especially for variable-star photometry in several filters. Mixed-model regression adds to normal multivariate regression certain “random effects”: categorical-variable terms that model and extract specific systematic errors such as image-to-image zero-point fluctuations (cirrus effect) or even errors in comp-star catalog magnitudes. In some contrast to the traditional approach of applying formulas to instrumental magnitudes in order to obtain estimated star magnitudes, the presented approach models measured data, then using the same model to predict the best estimated magnitudes of all check and target stars in the same large image set, typically all a night’s images. In the ideal case where no data are missing (e.g., no stars saturated or too faint), the approach is very similar to ensemble (multi-comp-star) photometry. However, the new workflow: (1) is robust to missing data points; (2) delivers plots that expose numerous types of systematic errors; and (3) is readily extensible to new terms, including isolating and removing the effects of imperfect image flat-fielding. Remaining challenges include (1) modeling hourly extinction-coefficient changes and (2) target-star magnitude uncertainty estimates that can be contaminated by large magnitude errors of comp stars from other sequences.