Toward a multidimensional analysis of transmission spectroscopy

Abstract

New-generation spectrographs dedicated to the study of exoplanetary atmospheres require a high accuracy in the atmospheric models to better interpret the input spectra. Thanks to space missions such as James Webb Space Telescope (JWST), ARIEL, and Twinkle, the observed spectra will indeed cover a large wavelength range from visible to mid-infrared with an higher precision compared to the old-generation instrumentation, revealing complex features coming from different regions of the atmosphere. For hot and ultra hot Jupiters (HJs and UHJs), the main source of complexity in the spectra comes from thermal and chemical differences between the day and the night sides. In this context, 1D plane parallel retrieval models of atmospheres may not be suitable to extract the complexity of such spectra. In addition, Bayesian frameworks are computationally intensive and prevent us from using complete 3D self-consistent models to retrieve exoplanetary atmospheres, and they require us to use simplified models to converge at a set of atmospheric parameters. We thus propose the TauREx 2D retrieval code, which uses 2D atmospheric models as a good compromise between computational cost and model accuracy to better infer exoplanetary atmospheric characteristics for the hottest planets. TauREx 2D uses a 2D parametrization across the limb which computes the transmission spectrum from an exoplanetary atmosphere assuming azimuthal symmetry. It also includes a thermal dissociation model of various species. We demonstrate that, given an input observation, TauREx 2D mitigates the biases between the retrieved atmospheric parameters and the real atmospheric parameters. We also show that having prior knowledge of the link between local temperature and composition is instrumental in inferring the temperature structure of the atmosphere. Finally, we apply such a model on a synthetic spectrum computed from a global climate model (GCM) simulation of WASP-121b and show how parameter biases can be removed when using 2D forward models across the limb.