Machine learning algorithms for understanding solar radial velocity: implications for exoplanet detection

Exoplanets, or extrasolar planets, are celestial bodies that orbit stars outside our own Solar System. The study of exoplanets holds immense significance for astrophysics and space science as it broadens our understanding of planetary formation, solar system dynamics, and the potential for life beyond Earth. The diverse range of exoplanets discovered to date - varying in size, composition, and orbital characteristics - offers invaluable insights into the complexities and commonalities of planetary systems across the Universe. One of the most consequential methods in the exoplanetary search is the Radial Velocity (RV) method, renowned for its efficacy in identifying Super-Earth planets orbiting bright stars and short-period Earth-mass planets. However, this method faces critical limitations in identifying solar-like stars and Earth twins, requiring a precision of approximately 10 cm/s - a demand almost an order of magnitude beyond current technological capabilities. Although the advent of next-generation spectrographs promises to attain such precision, the RV measurements remain susceptible to perturbations from stellar surface inhomogeneities like sunspots, introducing noise on the order of 1 m/s for the Sun. This level of interference renders the detection of Earth-like exoplanets unfeasible, even with advanced instrumentation. Consequently, investigating the effects of stellar structures on the RV signal is essential. This examination is critical since it holds the potential to refine the RV method’s sensitivity and pave the way for uncovering Earth-like exoplanets. Unlike distant celestial bodies, the Sun offers the unparalleled benefit of being observable both through high-resolution spatial datasets, such as disk-resolved intensitygrams, and also "as a star" via instruments like the High Accuracy Radial velocity Planet Searcher for the Northern Hemisphere (HARPS-N) solar telescope. This dual observation modality provides a fertile ground for research, allowing for a comprehensive understanding of Radial Velocity variabilities influenced by solar structures. Through the deployment of machine learning algorithms to synthesize these rich data sources, the research aims to establish more reliable predictive models for RV fluctuations and, thereby, refine the RV method for broader applications. Methodologically, a comprehensive suite of machine learning algorithms was deployed. Each algorithm was evaluated on its ability to identify correlations between high-precision RV measurements and various solar structures like sunspots, plage regions, and solar flares. In summary, this thesis provides a novel approach to overcoming the inherent limitations of the Radial Velocity method for exoplanet detection. By leveraging our unique observational access to the Sun and applying machine learning algorithms, the research presents a framework for understanding how the main solar structures affect radial velocity. The findings show that such a goal is possible, providing insights about the importance of each solar structure and bringing us a step closer to the long-sought goal of understanding more deeply the radial velocity method for discovering Earth-like planets around solar-like stars.