Use this URL to cite or link to this record in EThOS:
Title: Characterisation of exoplanetary atmospheres : from target selection to feature recognition using deep learning
Author: Zingales, Tiziano
ISNI:       0000 0004 7659 8758
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
On the day of writing this thesis, we know about 4000 exoplanets beyond the Solar System. We have a wide variety of known exoplanets, from very hot giant planets to cold Earths. Planetary detections, nevertheless, are not enough to thoroughly investigate the history and chemistry of the exoplanets. For this reason, atmospheric characterisation is becoming more critical than ever in exoplanetary science. In the next decade, many space missions such as JWST, Twinkle, ground-based instruments (ELT, TMT), and ARIEL, will study spectroscopically exoplanetary atmospheres and will help us examining more-in-depth planetary formation and dynamics. Today, the Hubble/WFC3 camera represents the state-of-art for transit spectroscopy. State of the art inverse models to interpret the observed exoplanetary spectra are based on Bayesian analysis, able to sample a sizeable parameter space and to converge on a possible real set of parameters that can explain the structure of exoplanetary spectra. In this thesis, I present the results obtained by applying the UCL Bayesian inverse model, TauREx, to the largest catalogue observed to date. I will demonstrate how it is possible to find water vapour in 16 out of 30 planets chosen from the WFC3 planetary dataset. Often the input spectra are too noisy to obtain a result statistically significant. For this reason, I will introduce the ADI (Atmospheric Detection Index) index, able to quantify the ``goodness'' and the significance of molecular detection. The use of complex atmospheric models on Bayesian analysis tools can require a prohibitive amount of time. For this reason, it is crucial to improve the analysis efficiency of complex atmospheres and accelerate their computations. To speed up the computation of atmospheric spectroscopic retrievals, I developed ExoGAN (Exoplanetary Generative Adversarial Network), a new-generation deep learning algorithm able to learn how to generate atmospheric spectra and retrieve the best set of parameters that can explain the observed spectrum. It consists of a deep convolutional generative adversarial network able to recognise molecular features, abundances and physical atmospheric parameters. Finally, after describing more ``traditional'' atmospheric retrieval tools, their optimisation using deep learning algorithm and their application to real data-sets (i.e. the HST/WFC3 camera), I introduce a possible target list of planet candidates for a space mission dedicated to transit spectroscopy: the ARIEL space mission. Target selection is a crucial task to optimally select the planets with different basic parameters to sample uniformly the whole orbital and physical parameters space. The generation of an optimal target list is highly dependent on the type of instrument, and it will critically influence the science return of the mission.
Supervisor: Tinetti, G. ; Waldmann, I. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available