The Planet Whisperer: Toward Characterizing Low-mass Planets from Doppler Surveys in the Presence of Stellar Activity

One way of detecting Earth-like planets around other stars is to look for very small shifts in the light signal, or spectrum, of a star that change when an exoplanet passes in front of its star. These spectra, however, are affected by many sources of noise that block these signal changes from being detected. Astronomers aim to develop ways of separating the signals of the exoplanets from the sources of noise. The hardest stellar noise to identify and remove is caused by spots that change with time on the star's surface. These investigators will look for differences between the signals of passing planets and signals of the surface spots on the star, and use these differences to create computer programs that can detect the differences. The investigators will create false data that can be used to test their programs. If successful, they will then apply these software tools to search data from space-based and ground-based telescopes, where they expect to find hundreds of Earth-like planets. This research serves the national interest by developing and applying new tools to detect planets that could harbor life. The techniques that are developed as part of this research will be included in classes and research at the university level. Astronomers are now searching for analogs of the Earth. Significant progress has been made in the development of more stable and higher resolution spectrometers with the potential to deliver precision sufficient to detect Earth analogs. However, stellar spectra are laced with quasi-periodic signals that arise from stellar photospheres: granulation, oscillations, spots and plage on rotating stars, and meridional flows. The most challenging of these stellar noise signals for exoplanet detection are spots and plage. The investigators propose to search systematically for differences between Doppler shifts from planets and the photospheric signals from spots and plage. They will use blind source separation techniques with principal component analysis, independent component analysis, dictionary learning algorithms, and multivariate Gaussian process modeling. A large library of simulated data will be created and used to verify the efficacy of the statistical tools for distinguishing Doppler shifts from photospheric noise sources. After validation, the most promising algorithms will be applied to real data from space-based and multiple ground-based telescopes. If signals of the level detected by the spaceborne Kepler telescope can be extracted in the presence of background noise sources, hundreds of Earth-like planets will be detected. The statistical techniques that are developed as part of this research will be incorporated into classes and research at the university level.