Statistics for Stars and Galaxies: techniques for non-parametric time series analysis and Bayesian inference in astronomy

点击翻译按钮获取中文摘要

Observational data in astronomy are different from experimental data because we cannot perform repeated experiments; although we can choose which objects to observe in the sky and which telescopes to use, ultimately the universe provides us with one sample. Thus, the statistical methods we employ are critical to the proper interpretation of these data. Astrostatistics is a relatively new interdisciplinary field that resides at the interface of astronomy and statistics and that seeks to create new knowledge in both fields. With astrostatistics, we can answer scientific questions about the universe while simultaneously discovering new statistical approaches for complicated, noisy data in the spatial and time domains. Interdisciplinary astrostatistics research groups have cropped up in both the United States and the UK in the past ten years. As new faculty at the University of Toronto, and jointly appointed between the Department of Astronomy & Astrophysics (DoAA, 51%) and the Department of Statistical Sciences (DoSS, 49%), I am in the unique position to lead the first astrostatistics research program in Canada. I envision a diverse Astrostatistics Research Team (ART) comprised of DoAA and DoSS undergraduate students, graduate students, and postdoctoral researchers who not only make contributions to the field of statistics but who also make groundbreaking discoveries in the field of astronomy. Through our research, the ART will train highly-qualified personnel with sought-after quantitative, technical, and qualitative skills. The long-term objectives of this research program are to (1) develop new statistical methods and model comparison techniques for studying the mass distribution of the Milky Way Galaxy, the Galactic Stellar Bulge, and Globular Clusters (GCs), and (2) rigorously test, validate, and build upon a new non-parametric time series analysis technique for studying the time-variability in stars. The interesting statistical challenges to overcome in both lines of research are that the data are incomplete and subject to significant measurement uncertainty, and that the physical models are non-linear. Our research will help us answer scientific questions about large and small systems in the universe, thereby advancing knowledge about dark matter, galactic evolution, and stellar evolution. Moreover, the new statistical methodologies and techniques will advance the field of statistics and have broad applications in other disciplines that perform Bayesian inference and time series analysis. With this in mind, the ART's research will be reproducible and open-source, so that the broader scientific community can benefit from our efforts. This program will also help prepare us for big data releases in the 2020s, including over 60TB of time series data from the Large Spectroscopic Survey Telescope (LSST) that will revolutionize the field of time-domain astronomy.