Modeling Techniques for Measuring Galaxy Properties in Multi-Epoch Surveys

Abstract

Data analysis methods have always been of critical importance for quantitative sciences. In astronomy, the increasing scale of current and future surveys is driving a trend towards a separation of the processes of low-level data reduction and higher-level scientific analysis. Algorithms and software responsible for the former are becoming increasingly complex, and at the same time more general - measurements will be used for a wide variety of scientific studies, and many of these cannot be anticipated in advance. On the other hand, increased sample sizes and the corresponding decrease in stochastic uncertainty puts greater importance on controlling systematic errors, which must happen for the most part at the lowest levels of data analysis. Astronomical measurement algorithms must improve in their handling of uncertainties as well, and hence must be designed with detailed knowledge of the requirements of different science goals. In this thesis, we advocate a Bayesian approach to survey data reduction as a whole, and focus specifically on the problem of modeling individual galaxies and stars. We present a Monte Carlo algorithm that can efficiently sample from the posterior probability for a flexible class of galaxy models, and propose a method for constructing and convolving these models using Gauss-Hermite ("shapelet") functions. These methods are designed to be efficient in a multi-epoch modeling ("multifit") sense, in which we compare a generative model to each exposure rather than combining the data from multiple exposures in advance. We also discuss how these methods are important for specific higher-level analyses - particularly weak gravitational lensing - as well as their interaction with the many other aspects of a survey reduction pipeline.