The Detection-vs-Retrieval Challenge: Titan as an Exoplanet

TL;DR

This work uses Titan’s exquisitely precise Cassini transmission spectra as a testbed for exoplanet atmospheric retrievals to quantify how pre-selection of molecular species biases inferred abundances, notably methane, by about $0.5$ dex. By exploring 25 molecular sets with a hierarchical retrieval framework and incorporating haze via a scale-height–coupled opacity, the study shows that overlapping hydrocarbon features create degeneracies that can masquerade as or obscure genuine detections. The authors advocate sensitivity analyses and chemistry-informed priors, and demonstrate a complementary path to constrain the dominant atmospheric constituent through the scale height, yielding a mean molecular weight of $\mu = 27.8\pm1.8$ amu and suggesting a background gas of $\mathrm{N}_2$ for Titan. Collectively, the results highlight fundamental limits of current exoplanet retrievals, the risk of over-interpreting detections, and the value of iterative, physics-informed approaches that couple retrievals to atmospheric-chemistry constraints and ancillary measurements.

Abstract

Cassini's observations of Titan's atmosphere are exemplary benchmarks for exoplanet atmospheric studies owing to (1) their precision and (2) our independent knowledge of Titan. Leveraging these observations, we perform retrievals (i.e., analyses) of Titan's transmission spectrum to investigate the strengths/limitations of exoplanet atmospheric retrievals with a particular focus on the underlying assumptions regarding the molecular species included in the retrieval. We find that multiple hydrocarbons can be ``retrieved'' depending on the selection made ahead of a retrieval. More importantly, we find that the estimates of other parameters such as the abundance of key absorbers like methane can be biased by $\sim$0.5 dex (by a factor of $\sim$3) due to such choices. This shows that beyond the possible misidentification of a molecular feature (e.g., current debate surrounding dimethyl sulfide, DMS, in K2-18 b), the implicit molecular detections made pre-retrieval to avoid retrieving for hundreds of molecules at a time can bias a large range of parameters. We thus recommend sensitivity analysis to assess the dependencies of atmospheric inferences on such selections in tandem with complementary information (e.g., chemistry models) to support any pre-retrieval selection. Finally, we introduce an independent path to constrain the dominant atmospheric constituent, even when lacking observable absorption feature (e.g., H$_2$ and N$_2$) through the scale height.

Figures (9)

• Figure 1: A family portrait of hydrocarbons (HCs) in Titan's transmission spectrum obtained with Cassini/VIMS. Best fit model (thick blue line), corresponding to the maximum-likelihood solution, is shown for Titan’s spectrum obtained during visit T10 (see \ref{['tab:VisitTable']}) from robinson2014, together with the individual contributions of the absorbing molecules (see legend) that comprise of the molecular combination case 25 (see \ref{['tab:MolecularSpeciesOG']}). The residuals, expressed in units of scaled observational uncertainty (see \ref{['eqn:likelihood']}), are shown in the bottom-right panel along with their distribution. An Anderson–Darling test indicates a deviation from Gaussianity at the 5% significance level.
• Figure 2: Atmospheric inferences are sensitive to the molecules retrieved for. Posterior distribution of the base atmospheric parameters for an ensemble of 25 "sets", associated with different selections of molecules retrieved (see \ref{['tab:MolecularSpeciesOG']}). The "truths" reported in \ref{['tab:MolecularSpecies']} are shown in the black dotted lines. Except for acetylene and carbon dioxide, the retrieved molecular values are generally consistent within 2$\sigma$ of previously reported values. Given the precision of the data, some extreme cases are highly unlikely. This is in particular the case of sets 3, 4, 5, and 6 that do not select a molecule with a sharp absorption feature at the center of the 3.3 $\mu$m band forcing large compensations on the abundance of methane and other parameters while still leading exceedingly large structures ($\sim$50 km vs $\sim$5 km otherwise) in the residuals (see \ref{['fig:goodModelbadModel']}). All other models lead to consistent fits and reveal a scatter of $\sim$0.5 dex among the mean retrieved abundances for absorbers such as methane.
• Figure 3: Spectra of different molecules may have many similarities, partially due to the presence of spectrally active shared functional groups sousaSilva2019zhan2021assessment. Overview of the absorption cross sections for HCs at 25$^{\circ}$C and 1.0 atm taken from PNNL PNNL_Sharpe_2004 and included in the cross sections part of the HITRAN2024 database XSC_HITRAN_2004. The y-axis provides the logarithm of intensity, with each absorption cross section offset for display purposes. A minimum intensity of $1.0 \times 10^{-21}$ cm$^{2}$/molecule has been applied for each molecule.
• Figure 4: Background gas identification via constraints on the scale height from transmission spectroscopy. Left: PPD of Titan's atmospheric scale height yielded by its transmission spectrum (consistent with the stratospheric value). Right Framework to yield robust abundances for terrestrial planets using iterative retrievals through identification of background gas.
• Figure 5: Weak absorption features muted during hazy epochs. Left: Titan's Transmission spectrum corresponding to two different visits (T10 and T53, see \ref{['tab:VisitTable']}). T53 is affected by hazes to such an extent that weak absorption features such as CO$_2$'s at 4.3 $\mu$m are mostly muted. Right: Retrieved CO$_{2}$ mixing ratios for T10 and T53 showing that in the latter case the detection of CO$_2$ is significantly less significant due to hazes muting its 4.3 $\mu$m band.