From grid models to machine learning
BORIS DOI
Subtitle
advancing atmospheric characterisation of brown dwarfs and exoplanets
Abstract
Thirty years ago, two groundbreaking discoveries—the brown dwarf companion Gliese 229 b (Nakajima et al., 1995) and the exoplanet 51 Pegasi b (Mayor and Queloz, 1995)—marked a significant turning point in substellar astrophysics. These discoveries have led to a remarkable increase in detections, confirming over 5900 exoplanets and 4000 brown dwarfs to date. With these large populations, research transitioned from an era focused on detection to one centred on characterisation. The properties of exoplanets and brown dwarfs encompass a wide range of extremes in temperature, mass, size, composition, and orbital configurations, posing challenges to our understanding of planetary and substellar evolution. Thanks to substantially improved precision of observational facilities, along with progress in computational technology, precise investigations of object-specific characteristics, as well as large population-level studies, can be conducted.
The research presented in this PhD thesis advances the atmospheric characterisation of brown dwarfs and exoplanets while simultaneously refining the underlying methodologies. This PhD thesis combines traditional Bayesian retrieval techniques, including nested sampling and MCMC, with supervised machine learning approaches such as random forests and develops a dedicated neural network for simulation-based inference, aiming to accelerate the retrieval process. Thereby, atmospheric model grids are critically examined, highlighting both their strengths and limitations. The PhD thesis also includes detailed case studies of individual substellar objects, such as the L7 dwarf VHS 1256b and the Saturn-mass gas giant WASP-39b, as well as broader population-level analyses across the brown dwarf spectral sequence. Each chapter addresses a distinct aspect of the complex, interconnected workflow required for robust atmospheric interpretation. By integrating physically motivated atmosphere models with machine learning techniques, this work establishes a foundation for efficient and scalable retrieval analysis—an essential step in preparing for the extensive volumes of high-fidelity data anticipated from existing and upcoming observational facilities.
The research presented in this PhD thesis advances the atmospheric characterisation of brown dwarfs and exoplanets while simultaneously refining the underlying methodologies. This PhD thesis combines traditional Bayesian retrieval techniques, including nested sampling and MCMC, with supervised machine learning approaches such as random forests and develops a dedicated neural network for simulation-based inference, aiming to accelerate the retrieval process. Thereby, atmospheric model grids are critically examined, highlighting both their strengths and limitations. The PhD thesis also includes detailed case studies of individual substellar objects, such as the L7 dwarf VHS 1256b and the Saturn-mass gas giant WASP-39b, as well as broader population-level analyses across the brown dwarf spectral sequence. Each chapter addresses a distinct aspect of the complex, interconnected workflow required for robust atmospheric interpretation. By integrating physically motivated atmosphere models with machine learning techniques, this work establishes a foundation for efficient and scalable retrieval analysis—an essential step in preparing for the extensive volumes of high-fidelity data anticipated from existing and upcoming observational facilities.
Date of Publication
2025
Year of graduation
2025
Theses Type
dissertation
Subject(s)
Language(s)
en
Author(s)
Faculty/Graduate School
Access(Rights)
open.access
Primary OA Publication
true