Atmosferische chemie van gasvormige exoplaneten: datagedreven modellering en statistische inferentie

The Universe is full of exoplanets, i.e. planets that orbit stars beyond our own Sun. The population of known exoplanets approaches 6000 and spans a wide range of masses, radii, and temperatures, including types of planets that are not found in our Solar System. With the advent of the James Webb Space Telescope (JWST), the field of exoplanet science has stepped into a new era in which we can study transiting exoplanets in unprecedented detail by characterizing their atmospheres. Via a technique called transmission spectroscopy, we can study atmospheric properties such as the chemical composition, climate dynamics, cloud coverage, and temperature structure, which in turn can inform us on the planet's formation mechanism, evolution history, and interior structure. Yet, because exoplanet atmospheres host a rich interplay of physical and chemical processes, it is crucial to develop a thorough theoretical understanding before interpreting observational data. In this thesis, we aim to better understand the chemistry of close-orbiting, gaseous exoplanet atmospheres. We do so through numerical modelling of the chemical composition, and bridge the gap between theory and observations by analysing transmission spectra taken by JWST. Stellar flares temporarily enhance the radiation flux incident on the planetary atmosphere, thereby affecting the efficiency of photochemical processes. Additionally, the climates of close-orbiting gas giants are dominated by a fast eastward jet stream along the equator (superrotation) that further complicates the distribution of chemical species. We consider the joint effect of stellar flaring and equatorial superrotation on the chemical composition and investigate if observations are affected. We find that stellar flares photochemically deplete CH4 and NH3 in the upper layers on the dayside and evening limb. Horizontal advection, induced by the eastward wind, enables photochemical production of C2H2 and HCN on the nightside and morning limb, leading to enhancements in their abundances of less than an order of magnitude. We predict that transit spectra are affected by several hundreds parts per million, which is similar to the precision of JWST. In a more realistic scenario of repeated flaring, the chemical composition can be permanently altered. The transit spectrum of super-Neptune WASP-107b was observed with the Mid-Infrared Instrument (MIRI) of JWST, which led to the detection of H2O, SO2, and silicate clouds. Surprisingly, no spectral features of CH4 were visible, indicating a low CH4 concentration despite theoretical predictions based on the expected temperature. Therefore, we construct models of the gas-phase chemistry to explain the inferred abundances of SO2 and CH4. We find that the strong SO2 features are explained by a metal enrichment (elements heavier than hydrogen and helium) in the elemental mixture, an extremely low planetary gravity, and the stellar radiation field of the host star. Furthermore, we outline several mechanisms that reduce the CH4 content to non-detectable levels, including carbon-depletion, deep quenching by strong vertical mixing, and/or an extremely hot internal heat flux that originates from tidal effects. Following the MIRI observations, WASP-107b was observed with additional instruments aboard JWST, revealing a rich chemical diversity of molecules such as SO2, H2O, CO2, NH3, CO, and even traces of CH4. The mixture of gases seems to suggest an atmosphere enriched in metals, carbon-deprived, inflated due to an extremely hot interior, and prone to disequilibrium chemistry effects such as vertical mixing and photochemistry. However, these findings vary among authors that apply different analysis techniques, ranging from computational inexpensive statistic fits (`retrievals`) with simplified and parametric models, to hybrid approaches that incorporate more complex forward models. To assess the reliability of WASP-107b's inferred atmospheric properties, we developed a hybrid retrieval method that performs Bayesian inference using a grid of 900 pre-computed atmospheric models spanning a range of metallicities (Z), carbon-to-oxygen ratios (C/O), intrinsic temperatures (T_int), and vertical mixing strengths (K_zz). In general, we find that parameters related to the elemental composition (Z, C/O) are well constrained, while others (T_int, K_zz) vary significantly with modelling choices and assumptions. In particular, we find that high-altitude clouds obscure the methane-rich layers, which prevents the retrievals from constraining the intrinsic temperature and vertical mixing strength. We conclude that this state-of-the-art retrieval can be useful for atmospheric characterization if the underlying grid of forward models is well understood.