Different Pipelines, Different Atmospheres?

JWST’s data are revolutionary, and exoplanet astronomers are learning to squeeze it for all it’s worth. Danger lies at the bleeding edges, though, and a recent study highlights how subtle assumptions can strongly influence final conclusions about a planet’s atmosphere.

Sniffing an atmosphere

As unprecedented as JWST data are, it is crucial to remember that after sniffing an exoplanet’s atmosphere, the telescope does not beam back a full inventory of the molecules it found. Instead, it sends back raw data from its distant perch at L2 that must be interpreted and analyzed by astronomers on the ground before any conclusions are drawn. Behind every discovery claim of a certain molecule in an exoplanet atmosphere is a laborious two-step process: “reduction” and “retrieval.”

A corner plot illustrating the results of an atmosphere retrieval, here computing different abundances and atmospheric profiles fit to a combination of the Tiberius output and Hubble data. [Constantinou et al. 2023]

The first of these, reduction, refers to the process of extracting the signal of interest from the raw data. For most exoplanet studies this means the transmission spectrum of the planet’s atmosphere, which is unfortunately convolved with other nuisance signals originating from quirks of the detector. To automate this extraction and cleaning, astronomers write software “pipelines” that turn the raw data downlinks into a clean final spectrum. Different pipelines often produce similar but slightly different outputs depending on the design choices underpinning each.

Once raw data are pushed through a pipeline and the spectrum is extracted, astronomers move on to the second step: retrieval. Here, the goal is to explain the spectrum by using complex models that simulate a planet’s atmosphere to match what was observed.

Products Depend on Pipelines

Two reduction pipelines popular in the JWST literature are named Tiberius and Eureka. When fed the exact same raw data, they return spectra that look qualitatively similar but differ slightly at each wavelength. Recently, a collaboration led by Savvas Constantinou (University of Cambridge) decided to investigate the implications of these subtle differences: they passed each pipeline an identical copy of observations of the “hot Saturn” exoplanet named WASP-39 b, then ran retrievals on each output spectrum to check how the final atmospheric profiles depended on the choice of pipeline.

Differences in retrieved parameters when using Tiberius and Eureka. Appending Hubble data to each lessened but did not fully resolve the discrepancies between them. [Constantinou et al. 2023]

They found that while the analyses agreed on the largest signals (both successfully recovered the much-lauded detections of CO2 and SO2), they differed in statistically meaningful ways when estimating more subtle parameters. For example, the inferred mixing ratio of most species differed by more than 1σ, and in some cases, one pipeline led to a molecule detection while the other concluded it wasn’t present. Although adding in older data from the Hubble Space Telescope helped reconcile the two somewhat, some serious differences remained even then.

The authors concluded that while there is no cause for concern when it comes to interpreting the most obvious whopping signals like the >15σ CO2 absorption, astronomers are going to need to take care when interpreting weaker signals. With many more JWST transit observations to come, the community will have to double check that they don’t miss any molecules, and that they ones they claim float in exo-airs are really there and aren’t just figments of our reductions.


“Early Insights for Atmospheric Retrievals of Exoplanets Using JWST Transit Spectroscopy,” Savvas Constantinou et al 2023 ApJL 943 L10. doi:10.3847/2041-8213/acaead