Abstract
Atmospheric retrieval is the inverse modeling method where atmospheric properties are constrained based on measured spectra. Due to the low signal-to-noise ratios of exoplanet observations, exoplanetary retrieval codes pair a radiative transfer (RT) simulator with a Bayesian statistical framework in order to characterize the distribution of atmospheric parameters that could explain the observations (the posterior distribution). This requires on the order of 106 RT model evaluations, which requires hours to days of compute time depending on model complexity. In this work, I investigate atmospheric retrieval methods and apply them to observations of hot Jupiters. Chapter 2 presents a set of RT and retrieval tests to validate the Bayesian Atmospheric Radiative Transfer (BART) retrieval code and applies BART to the emission spectrum of HD 189733 b. Chapter 3 investigates the dayside atmosphere of WASP-12b and resolves a tension in the literature over its composition. Chapter 4 introduces a machine learning direct retrieval framework which spawns virtual machines, generates spectra, trains neural networks, and performs atmospheric retrievals using trained neural networks. Chapter 5 builds on this and presents a machine learning indirect retrieval method, where the retrieval is performed using a neural network surrogate model for RT within a Bayesian framework, and compares it with BART. Chapter 6 utilizes the neural network surrogate modeling approach for thermochemical equilibrium chemistry models and compares it with other equilibrium estimation methods. Appendices address retrieval errors induced by choice of wavenumber gridding for opacity-sampling RT schemes, neural network model selection, the effects of data set size on neural network training, and the accuracy of Bayesian frameworks used for atmospheric retrieval.
Notes
If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu
Graduation Date
2022
Semester
Summer
Advisor
Harrington, Joseph
Degree
Doctor of Philosophy (Ph.D.)
College
College of Sciences
Department
Physics
Degree Program
Physics; Planetary Sciences
Identifier
CFE0009189; DP0026785
URL
https://purls.library.ucf.edu/go/DP0026785
Language
English
Release Date
August 2022
Length of Campus-only Access
None
Access Status
Doctoral Dissertation (Open Access)
STARS Citation
Himes, Michael, "Atmospheric Retrieval: Bayesian Methods, Machine Learning, and Application to Exoplanets" (2022). Electronic Theses and Dissertations, 2020-2023. 1218.
https://stars.library.ucf.edu/etd2020/1218