The effect of JWST/NIRSpec data reduction on the retrieval of WASP-39b atmospheric properties

TL;DR

JWST/NIRSpec transit spectra enable detailed atmospheric constraints but data-reduction systematics and saturation handling can bias retrievals. The study analyzes six independent reductions of WASP-39b, comparing full-range data with saturated-region-excluded data ($0.69-1.91 \,\mu\mathrm{m}$) and tests three cloud-extinction models using nested-sampling retrievals with MultiNest and PSG forward models. They find that inter-pipeline differences can cause variations in $T_{iso}$, molecular abundances (e.g., H$_2$O, CO$_2$, CO) and cloud opacity, frequently exceeding an order of magnitude; excluding the saturated region reduces dispersion but increases degeneracy, and Bayesian evidence favours non-flat aerosol extinction, depending on calibration. These results underscore the need for robust, homogeneous calibration in JWST exoplanet studies and confirm that JWST data retain sensitivity to aerosol spectral behavior when data reductions are carefully managed.

Abstract

The JWST provides exoplanet transit observations with unprecedented spectral coverage, enabling detailed atmospheric characterisation. However, systematics introduced during data reduction can lead to small but significant uncertainties that propagate into atmospheric retrievals, making it essential to assess their impact on inferred properties. We aim to quantify the impact of different JWST/NIRSpec PRISM data-reduction processes as well as the relevance of saturation on the retrieved atmospheric properties of WASP-39b. We also assess whether or not these biases are comparable to those introduced by assumptions made in atmospheric modelling, particularly in the treatment of aerosol extinction. We perform nested-sampling Bayesian retrievals using MultiNest and forward models generated with the Planetary Spectrum Generator. Six independently reduced spectra are analysed, considering both the full wavelength range and versions excluding the saturated region. We further test the effect of including three different cloud-opacity parameterisations. Differences among JWST/NIRSpec data-reduction pipelines lead to substantial variations in retrieved atmospheric properties of WASP-39b, often exceeding one order of magnitude, comparable to uncertainties from modelling assumptions. Excluding the saturated region reduces inter-pipeline dispersion but increases parameter degeneracies. This highlights the need for robust and homogeneous calibration strategies. The results also confirm that JWST data possess the sensitivity required to probe aerosol spectral behaviour, although such constraints remain strongly dependent on the adopted data-reduction strategy.

Figures (14)

• Figure 1: Representation of the different spectral data used for the purpose of this work. The gray shades in the 0.63 - 2.06 micron range are related to the level of saturation: for the lightest region, only one group per integration was affected by saturation, while for the darkest region, up to four of the five groups were affected.
• Figure 2: Comparison of the spectra with the FIREFLy data reduction as a reference and their respective residuals. Coloured regions at the residuals represent error bar propagation. Grey regions show the saturated spectral range (0.63 - 2.06 $\mu$m) as in Fig. \ref{['FigEspecs']}.
• Figure 3: Comparison of the retrieval results obtained when using the different spectra as input (shown in different colours). See section \ref{['sec:Data']} for a full reference of the spectra. The gray values represent the reference values from Faedi2011. The planetary radius at 100 mbar, the mean atmospheric molecular weight and the aerosol opacity at 1 $\mu$m were computed from actual outputs and not directly retrieved.
• Figure 4: Comparison of the results obtained when using the different spectra in retrievals (shown in different colours). See section \ref{['sec:Data']} for a full reference of the spectra. The filled points correspond to the spectra covering the full spectra range and the unfilled ones to the spectra with the saturated spectral range removed. The gray values represent the reference values from Faedi2011. The planetary radius at 100 mbar, the mean atmospheric molecular weight and the aerosols opacity at 1 $\mu$m were computed from actual outputs and not directly retrieved.
• Figure 5: Partial corner plot extracted from Figure \ref{['FigCornerPlotEureka']}, using Eureka spectrum as the input for the retrievals. Black areas are used for the results including the whole spectral range, while red is used for those excluding the saturated 0.69 - 1.91 $\mu$m range. Temperature shows negative correlations with molecular abundances, while the abundances of $\rm{H_2O}$, $\rm{CO_2}$ and $\rm{CO}$ are positively correlated with each other.