Over the course of roughly four years, NASA’s Kepler Space Telescope (launched in 2009) monitored the brightness ~ 200,000 stars in its field of view. Using instruments sensitive enough to detect Earth-sized planets in year-long orbits around Sun-like stars, Kepler aimed to constrain the occurrence rate of such objects, otherwise known as eta-Earth. Although the Kepler mission fell short of detecting a true Earth-Sun analog (due to higher amounts of noise than originally anticipated), it has provided humanity with important insight into (1) how the astronomy community should interpret the completeness and reliability of exoplanet studies and (2) how uncertainties in the observed sizes of exoplanets impacts our ability to constrain eta-Earth.
Stars monitored by Kepler that show periodic changes in brightness indicative of a transiting exoplanet are known as Kepler Objects of Interest (KOIs). Data from Kepler undergoes a series of processing steps before being analyzed by astronomers to remove any effects caused by instrumental systematics. The resulting data should then ideally only contain systematics originating from the star itself (sunspots, stellar flares, oscillations in brightness, etc.) in addition to signatures caused by exoplanets or stellar binaries; however, instrumental systematics can also persist. In some cases, these forms of noise can even result in “false alarm” transit detections.
My M.Sc. research focused on modeling the possible impact of stellar and instrumental noise on Kepler data to assess the probability of false alarm detections. To accomplish this, I developed a method for modeling KOI light curve data as solely noise (false alarm) or as noise superimposed on a transit curve (real exoplanet). I then used Bayesian statistical methods to determine the probability that a given KOI is a transiting planetary candidate or a false alarm. This enables astronomers to better understand the contamination rate of false alarms in KOI samples; a key component in estimating catalog reliability.
My models incorporate both “white noise” (with constant intensity) and “correlated noise” (where the noise levels between pixels or over time are connected). Simultaneous modelling of white and correlated noise provides tighter constraints on transit model parameters, thereby improving the characterization of planetary candidates. This allows for better estimates of exoplanet radii, which in turn refines reliability measurements and eta-Earth estimations.
Having prototyped this novel approach on a select subset of KOIs, including potential Earth-Sun and Venus-Sun analogs, my Ph.D. project will leverage a simulation-based inference machine learning architecture to facilitate efficient Bayesian reliability assessment of exoplanet transit signatures on catalog-wide data sets (Kepler, K2, TESS, etc.).
Michael completed his MSc at Bishop’s between 2021 and 2023, under the supervision of Prof. Jason Rowe. His thesis will be available soon.