Knobs and dials of retrieving JWST transmission spectra

Context. Since the launch of the James Webb Space Telescope (JWST), observations of exoplanetary atmospheres have experienced a revolution in data quality. As atmospheric parameter inferences heavily depend on the underlying dataset, a re-evaluation of current methodologies is warranted to assess the reliability of these results. Aims. We investigate the impact of variations in input spectra on atmospheric retrievals for the hot Jupiter WASP-39 b using JWST transit data. Specifically, we analyse the reliability of parameter estimation results from random perturbations of the underlying spectrum, and their sensitivity to the use of three transmission spectra derived from the same observational data. Methods. Using the NIRSpec PRISM observation from a single transit of WASP-39 b, we performed retrievals with the TauREx framework. As an input baseline, we used a transmission spectrum derived in our work using the Eureka! data reduction pipeline. To investigate the reliability of these retrieval results, we analysed the behaviour of parameter posterior distributions under deviations of this spectrum. To mimic random noise, we performed a set of retrievals on scattered instances of the spectrum produced in this work. We compared this to differences resulting from retrievals based on existing spectra reduced from the same raw observation. Results. Our analysis identifies three types of parameter posterior distributions: (1) stable, Gaussian distributions for species constrained across the entire spectrum (e.g. H2O and CO2); (2) uniform posteriors with upper bounds for species with weak constraints (e.g. CO and CH4); and (3) unstable, heavy-tailed posteriors for species constrained only by minor spectrum features (e.g. SO2 and C2H2). We find that other parameters, such as the planetary radius and pressure-temperature profile, are stable under spectral perturbations. Conclusions. Parameter posterior distributions are different for atmospheric retrievals performed on independently reduced transmission spectra derived from the same raw data. This makes a robust interpretation difficult, particularly for skewed distributions. Based on this, we advocate for the careful assessment and selection of credible interval sizes to reflect this.